forestcarbonpartnership.org · LIBERIA REDD+ REFERENCE SCENARIO DELIVERABLE 4 – FINAL REPORT 2...

Development of Liberia’s REDD+ Reference

Level

Final Report for Republic of Liberia Forest

Development Authority

Katherine Goslee, Sarah Walker, Edward Mitchard, Alex Grais, Mike

Netzer, Kevin Brown, Lara Murray, Jessica Donovan, Peter Mulbah

October 2016

LIBERIA REDD+ REFERENCE SCENARIO

DELIVERABLE 4 – FINAL REPORT

2

TABLE OF CONTENTS Executive Summary ............................................................................................................................ 4

1. Introduction ........................................................................................................................... 14

1.1 Liberia’s Unique National Circumstances .......................................................................................... 19 1.2 A Note on Available Data .................................................................................................................. 24 2. Defining the Scope of the REL/RL ............................................................................................. 25

2.1 International Guidance on RL Development ..................................................................................... 25 2.2 Forest Definition ................................................................................................................................ 27 2.3 Scope ................................................................................................................................................. 28 2.4 Reference Period ............................................................................................................................... 29 2.5 Scale .................................................................................................................................................. 30 2.6 Pools and gases ................................................................................................................................. 31 3. Activity Data ........................................................................................................................... 31

3.1 Evaluation of existing spatial datasets .............................................................................................. 32 3.2 Deforestation Rate Estimation .......................................................................................................... 35 3.3 Characterization of Deforestation Trends ......................................................................................... 40 3.4 Deforestation Activity Data development ........................................................................................ 44 4. Emission Factors ..................................................................................................................... 46

4.1 Forest biomass carbon stocks ........................................................................................................... 46 4.2 Post deforestation biomass carbon stocks ....................................................................................... 49 4.3 Soil carbon stocks .............................................................................................................................. 50 4.4 Emission factors ................................................................................................................................ 52 5. Reference Level Development ................................................................................................. 53

5.1 Historical Emissions ........................................................................................................................... 53 5.2 Projecting Future Emissions without REDD ...................................................................................... 55 5.3 Adjusting for national circumstances ................................................................................................ 56 5.4 Reference Emission Level .................................................................................................................. 60 6. Next Steps and Recommendations .......................................................................................... 60

6.1 Submitting the proposed REL ............................................................................................................ 60 6.2 Future improvements ....................................................................................................................... 61 7. References .............................................................................................................................. 65

Appendix A: Available land cover, land cover change, and raw satellite data ...................................... 68

Appendix B: Field assessment of spatial dataset accuracy .................................................................. 73

Appendix C: Creation of Land Cover Change products ........................................................................ 81

Appendix D: Existing biomass data, allometric equations and biomass expansion factors ................... 89

Appendix E: Recommendations for Improving Deforestation Activity Data for Liberia ...................... 100

Appendix F: Mapping Land Use with Remote Sensing ...................................................................... 115



3

ANNEX 1: Forest Degradation ......................................................................................................... 117

ANNEX 2: Economic analysis of baseline drivers of land use change in Liberia .................................. 148

ANNEX 3: Capacity Building Strategy ............................................................................................... 175



4

EXECUTIVE SUMMARY

Reducing Emission from Deforestation and Forest Degradation, and enhancing forest carbon stocks

(REDD+) in developing countries is a valuable mechanism for countries that aim to mitigate the impacts of

climate change by cutting carbon dioxide emissions originating from the destruction of forests. In Liberia,

REDD+ is viewed as an opportunity and a viable source of sustainable finance for investment in forest

management, forest conservation, and forest restoration. Additionally, Liberia recognizes the multiple

benefits REDD+ may offer, including but not limited to biodiversity conservation, watershed management,

enhanced resilient capacity and poverty reduction. The performance based climate financing REDD+ may

provide Liberia with is seen as a viable opportunity that will enable Liberia to financially benefit from its

forests without degrading them.

Vast tropical forests cover nearly half of Liberia’s land mass, which are essential to the livelihoods of

Liberia’s peoples as well as the health of its ecosystems. While Liberia’s forests have historically been

subject to exploitation, it has had relatively low deforestation rates compared to many of its neighbors.

In fact, according to some estimates, the country contains over half of West Africa’s remaining rainforests1.

Nevertheless, given the fact that over half of Liberia’s forest land has been allocated either as commercial

concessions or is designated for conservation as protected areas, the potential for significant land use

change and associated emissions in Liberia should be considered high. To date, most of the concession

areas have yet to be developed and protected areas are not yet well established, leaving the future of

Liberia’s forests unknown.

The Government of Liberia has already made strides in forest governance as well as in other particular

requisites for REDD+. Advancements include the creation and mobilization of necessary institutions and

frameworks with the mandate of supporting the development of REDD+ in the country, including the

National Climate Change Steering Committee (NCCSC), REDD+ Technical Working Group (RTWG) and

REDD+ Implementation Unit (RIU). Additionally, the Government has enacted laws and policies to

advance REDD+ strategy and submitted a Readiness Preparation Proposal (R-PP) to the Forest Carbon

Partnership Facility (FCPF), approved in March 2012. Liberia’s Voluntary Partnership Agreement (VPA) for

Forest Law Enforcement, Governance, and Trade with the European Union further supports transparent

and sustainable forest management. In addition, the partnership between the Governments of Liberia and

Norway announced in September 2014 further provides the financial and technical foundation for

successful implementation of REDD+ in Liberia. Liberia has also received assistance from the World Bank’s

1 http://www.euflegt.efi.int/liberia

http://www.euflegt.efi.int/liberia



5

Forest Carbon Partnership Facility (FCPF) to develop and apply strategies to reduce emissions from

deforestation and forest degradation.

This report describes a recommended REDD+ Reference Emission Level for Liberia. Sections 1-5 represent

information directly related to REL development needed for submission of a proposed REDD+ Reference

Level to the UNFCCC or the FCPF. Section 6 provides recommendations for next steps and improvements.

International Guidance on RL Development

There are two main sources of guidance on the development of a REDD+ Reference Level: the United

Nations Framework Convention on Climate Change (UNFCCC) and the World Bank Carbon Fund’s

Methodological Framework. The UNFCCC provides general recommendations for the development of an

internationally acceptable Reference Level, while the Carbon Fund’s Methodological Framework provides

more explicit requirements for receiving funding under the Carbon Fund. Both refer to accounting

methods described by the Intergovernmental Panel on Climate Change (IPCC).

UNFCCC Conference of Parties (COP) decisions contain modalities that guide the development of forest

reference levels, particularly decision 12/CP.17 and its Annex. According to these modalities, Parties must

be transparent in establishing RLs, taking into account historical data and, if appropriate, adjusting for

national circumstances2. While forest RLs can be developed sub-nationally as an interim measure while

transitioning to a national scale, Liberia has chosen to develop its RL at a national scale. A step-wise

approach may be used, allowing Parties to improve the forest RL by incorporating better data,

methodologies and additional pools, if appropriate. Forest RLs are expressed in units of tons of CO2

equivalent per year and must maintain consistency with a country’s greenhouse gas inventory (according

to 12/CP.17, Paragraph 8). In response to the guidelines for submissions of information on RLs provided

in decision 12/CP.17, a summary of Liberia’s decisions on these modalities is given in Table ES-1.

Table ES-1. UNFCCC modalities relevant for Liberia's national Reference Level

Reference to

Guideline

Description Liberia’s Proposal

Decision

12/CP.17

Paragraph 10

Allows for a step-wise

approach

REL is at national scale, and includes all drivers of deforestation

Degradation will be added as a stepwise improvement, as additional data become available.

2 Decision 4/CP.15, paragraph 7.



6

Decision

12/CP.17 Annex,

paragraph (c)

Pools and gases included Pools: (activity specific) - Aboveground and belowground

biomass

- Dead wood

- Litter

- Soil carbon

Gases: - Include CO2

Decision

12/CP.17 Annex,

paragraph (c)

Activities included Include deforestation caused by agriculture, mining, forestry infrastructure, and other infrastructure

Other activities will be included in step-wise improvements of the REL

Decision

12/CP.17 Annex,

paragraph (d)

Definition of forest used is

same as that used in national

GHG inventory

Minimum tree cover: 30%

Minimum height: 5 m

Minimum area: 1 ha

Decision

12/CP.17 Annex

The information should be

guided by the most recent IPCC

guidance and guidelines,

All data are gathered using best practices and integrated to estimate emissions using IPCC 2003 and 2006 guidelines3

Where country specific data are not available, they will be developed

Decision

12/CP.17 II.

Paragraph 9

To submit information and

rationale on the development

of forest RLs/RELs, including

details of national

circumstances and on how the

national circumstances were

considered

Liberia proposes an upward adjustment to its reference level, as due to national circumstances, historical emissions likely do not accurately reflect future emissions. However, additional data are needed to identify a justifiable number for adjustment.

In addition to the decisions described in Table ES-1, it is necessary to establish a Reference Period, the

period from which data on past changes in forest area are established, analyzed, and projected into the

future. This is used to determine the average annual level of emissions against which future years are

3 The two IPCC reports used are the IPCC 2003 Good Practice Guidance for the LULUCF sector (IPCC 2003 GPG) and the IPCC 2006 Guidelines for National GHG Inventories, Volume 4 AFOLU (IPCC 2006 AFOLU)



7

compared. There are a number of factors that must be considered in determining an appropriate

reference period, though it is dictated in part by available data. In the case of Liberia, there are reliable

data available on forest loss from 2000 through 2014.

Liberia’s national circumstances have significantly influenced historical deforestation rates, including a

decade of civil war ending in 2003, timber sanctions enacted between 2003 and 2006, economic decline

through 2005, and the Ebola outbreak in 2014. These events have heavily influenced historical rates of

deforestation, and as the country’s economy begins to improve, land use and land cover change patterns

will likely shift. As such, based on the recent history of Liberia, a reference period of 2005-2014 is

recommended.

Estimating land cover change

Land cover change was estimated using three main sets of spatial data – the 2014 Land Cover Map

produced by Metria/Geoville (2016), a 2000 Percent Forest Canopy Cover map (Hansen et al 2013), and a

Forest Loss product for 2000-2014 (Hansen et al 2013). The 2014 Liberia Land Cover Map developed by

Metria and Geoville (2106) is a high resolution (10m) map based on RapidEye and Landsat 8 data. This

map serves as the best and most recent forest classification for Liberia, and will therefore be used as the

minimum level of stratification of forest cover in the Reference Level.

In 2013, Hansen et al published a set of global spatial products that span 2000 to 2013. This includes a

global 30-meter resolution ‘2000 Percent Forest Canopy Cover’ map and a 30-meter resolution ‘Annual

Forest Loss’ product produced annually from 2000 onwards. The Hansen et al (2013) ‘2000 Percent Forest

Canopy Cover’ precisely matches the ‘Annual Forest Loss’ product produced from 2000 onwards (Hansen

et al 2013), which can be used to create a series of land cover change products and thus create annual

deforestation activity data. The Annual Forest Loss product analyzes all available Landsat imagery and

combines training data from across the planet to estimate forest loss from (currently) 2000 to 2014

(Hansen et al 2013).

These data have been freely released by the University of Maryland4 and will be annually updated. The

datasets have been widely publicized through Global Forest Watch5 and the use of such datasets is

recommended in the Global Forest Observations Initiative’s Methods and Guidance Document

“Integrating Remote-sensing and Ground-based Observations for Estimation of Emissions and Removals

of Greenhouse Gases in Forests”6 (referred to as ‘MGD’). GFOI MGD explicitly encourages the use of this

4 http://www.earthenginepartners.appspot.com/science-2013-global-forest 5 http://www.globalforestwatch.org/ 6 http://www.gfoi.org/methods-guidance/

http://www.gfoi.org/methods-guidance/



8

dataset and now has a module advising on how this dataset can be used in the development of country-

level reference levels7. However, the MGD and various research indicates that the Hansen et al (2013) can

have significant local biases in its canopy cover estimates, and thus need local correction before being

used for Liberia.

Based on this evaluation, it was concluded that the Hansen et al dataset provides the highest resolution

data available to identify forest loss on an annual basis. As recommended by the GFOI MGD, the data was

first processed to create a ‘Liberia-Corrected 2000 Percent Canopy Cover’ map, stratified by forest class

(30-80% Canopy Cover; >80% Canopy Cover), using a combination of information from the Metria/Geoville

map and the published Hansen et al (2013). This dataset was used to produce annual forest strata maps

and annual deforestation estimates for the reference period, 2005-2014 (Table ES-2). These were based

on the Hansen 2000 canopy cover map, corrected for Liberia using the 2014 Metria Geoville classifications,

with annual forest loss calculated from that date onwards.

In general, deforestation appears to follow two patterns: large clearings associated with the creation of

plantations, visible as five clusters each 20-50 km from the coast stretching down the whole coastline, and

a network of tiny clearings located broadly throughout the country, though focused around roads and

settlements. The small clearings are likely associated with agriculture and logging, and are relatively evenly

spread throughout the time period, whereas the plantation clearings seem to be focused from 2011

onwards.

7 http://www.gfoi.org/wp-content/uploads/2015/03/MGDModule2_Use-of-Global-Data-Sets.pdf



9

Table ES-2. Annual rate of forest loss over reference period 2005-2014, by forest cover class

Year deforested Forest >80% Forest 30-80% Combined forest loss

percent loss

2005 0.06% 0.20% 0.08%

2006 0.19% 0.73% 0.26%

2007 0.26% 0.93% 0.35%

2008 0.20% 0.82% 0.28%

2009 0.50% 1.71% 0.67%

2010 0.16% 0.42% 0.19%

2011 0.24% 0.70% 0.30%

2012 0.48% 1.34% 0.60%

2013 0.85% 2.48% 1.07%

2014 0.71% 1.92% 0.87%

Average annual

loss 0.36% 1.07% 0.46%

Activity data were developed to reflect the magnitude of conversion from each forest class to six post-

deforestation land uses: shifting cultivation, oil palm plantations, rubber plantations, non-forest mixed

vegetation, mining, and settlements.

Options for additional stratification of Liberia’s forests were also examined, indicating that it would be

useful for stratification based on concession areas, as these areas experience a higher rate of deforestation

than the rest of the country. It would also be beneficial to stratify by road access, as most deforestation

occurred within 2 kilometers of roads, and focusing sampling efforts in these areas would increase

efficiency.

While the current focus of the REL is on deforestation, emissions from forest degradation are likely to

make a substantial contribution to land use emissions in Liberia, and so two possible methods for

estimating degradation were examined. Due to a lack of suitable data for either method, the results from

the two varied widely, indicating a need for more reliable data, and a future step-wise addition of

degradation in the REL, as such data become available.

Emission Factors

Various sources of data may be used to estimate forest biomass and develop emission factors, including

carbon measurement inventories, forest inventories, research studies, and global default values. Currently



10

there are limited data available for Liberia, and emission factors were developed using existing global

datasets. Three available pantropical maps of aboveground biomass were assessed for appropriateness in

producing carbon stocks for each of Liberia’s forest cover classes identified in the Metria/GeoVille map:

Avitabile et al. (2015), Saatchi et al. (2011), and Baccini et al. (2012). The carbon stocks for aboveground

biomass from Avitabile et al. (2015) match most closely with existing data from Liberia and neighboring

countries, and the differences between forest classes are most realistic. However, for this Reference Level,

we have used carbon stocks from Baccini et al (2012) to develop provisional emission factors, because

they provide more conservative estimates and will result in a lower reference level. Additional, smaller

carbon pools – belowground biomass, leaf litter, and deadwood – are included using default IPCC values.

Post-deforestation biomass carbon stocks have been estimated based on land-use, but these should be

improved where possible, to be based on country-specific data. Soil carbon stocks were sourced from the

Harmonized World Soil Database, with the amount of soil carbon emitted as CO2e estimated as a function

of land-use practices that follow forest loss, according to IPCC guidelines. Provisional deforestation

emission factors are shown in Table ES-3.

Table ES-3. Deforestation emission factors by forest class and post-deforestation land use, using forest carbon stock data from

Baccini et al (2012)

Stratum EF (t CO2e ha-1)

Shifting

cultivation

Oil palm

plantation

Rubber

plantation

Non-forest

mixed

vegetation

Mining Settlement

Forest > 80%

Canopy Cover 356.9 390.4 225.4 528.8 553.7 500.4

Forest 30-80%

Canopy Cover 303.2 446.3 171.3 472.9 497.5 446.3



11

Historical Emissions

Historical emissions were estimated as the product of activity data and emission factors (Table ES-4).

Table ES-4. Historical emission estimates for Reference Period, by forest class and post-deforestation land use, based on

emission factors from Baccini et al dataset.

Sum of All Forests

Shifting

Cultivation

Oil palm

Plantations

Rubber

Plantations

Non-forest

mixed

vegetation

Mines Settlement Total

Year Emissions (t CO2e)

2005 853,775 324,969 17,344 816,644 212,855 2,535 2,228,122

2006 2,715,457 1,048,722 55,763 2,613,782 680,629 8,019 7,122,372

2007 3,634,450 1,393,931 74,250 3,487,838 908,643 10,762 9,509,874

2008 2,871,385 1,115,441 59,221 2,770,917 721,273 8,461 7,546,699

2009 6,895,987 2,630,404 140,312 6,602,159 1,720,588 20,462 18,009,913

2010 1,978,554 739,695 39,664 1,877,985 490,057 5,915 5,131,871

2011 3,096,437 1,165,064 62,369 2,947,113 768,726 9,235 8,048,944

2012 6,158,690 2,307,277 123,655 5,850,861 1,526,567 18,397 15,985,448

2013 10,950,159 4,114,790 220,350 10,416,335 2,717,229 32,674 28,451,538

2014 8,856,326 3,304,035 177,270 8,398,617 2,191,904 26,497 22,954,649

Recommended Reference Level

The average historical emissions over the reference period 2005-2014 are 12,498,943 t CO2e/year based

on Baccini et al (2015) data and estimates of land use change. This represents a Reference Emission Level,

without adjustments for national circumstances.

The World Bank Carbon Fund allows the Reference Level to be based on an adjustment of the average

annual historical emissions over the Reference Period, not to exceed 0.1% of carbon stocks, if a country

can demonstrate that historical emissions from land use change do not adequately represent anticipated

increases in future emissions. In order to use an adjusted Reference Level, it is necessary to develop a

defensible number that can be used to adjust the historical average emissions. This requires additional

information on planned or expected development in the country. An initial justification is based on the

draft Land Use Analysis report (LTS, 2016) and indicates that expected land use change without REDD

activities would result in emissions exceeding the 0.1% cap. Therefore, an initial adjusted Reference



12

Emission Level could be proposed at 15,343576 t CO2e/yr. It is important to note, however, that it is not

known whether the FCPF would allow Liberia to use an adjusted Reference Level.

Uncertainties

The FCPF Carbon Fund Methodological Framework Section 3.2 requires that sources of uncertainty in

Reference Level setting are systematically identified and assessed; and further managed/reduced to a

feasible extent. This applies to all potential sources of uncertainties in both Activity Data and Emission

Factors. A lack of ground data and local layers led to the use of global/pantropical and IPCC Default values

for various components of the Activity Data and Emission Factor calculations: this not only increases

uncertainty, but also reduces our ability to estimate these uncertainties. We consistently chose data and

approaches that reduced uncertainties and potential biases as much as possible, but nonetheless

systematic as well as random errors will inevitably remain, and their degree will be hard to estimate.

We conducted a field campaign in order to estimate the accuracy of the 2014 landcover map and give 95%

confidence intervals for the forest cover strata for 2014, following the Olofsson et al. (2013)8 method as

recommended by the GFOI MGD6. Unfortunately, it was not possible to set up the number of plots advised

by this method, nor place these in a stratified random manner across the country, and thus the 95%

confidence intervals themselves have high uncertainty. Further, no suitable ground data on past landcover

or the location of deforestation/degradation were available, despite extensive discussions with

stakeholders in country and searches of the published and grey literature. This meant that while we could

produce confidence intervals for the 2014 strata, it was not possible to backdate these through time using

national data.

We did provide an estimate of the 95% confidence intervals for the deforestation data, and thus the

minimum and maximum ranges of strata, using the generic tropical validation data provided by Hansen et

al. (2013). These data are not specific to Liberia, and not stratified by forest type, and thus only provide

an indication of the potential errors caused by a combination of errors of commission (where change is

recorded when in fact no change occurred), and errors of omission (where a change is missed).

Uncertainty estimates for Emissions Factors are impossible to estimate without any ground data, and thus

error estimates for this side of the RL/REL calculations will require ground plots to be set up, ideally as part

of a full National Forest Inventory.

8 Olofsson, P., Foody, G.M., Stehman, S.V., & Woodcock, C.E. (2013). Making better use of accuracy data in land change studies: estimating accuracy and area and quantifying uncertainty using stratified estimation. Remote Sensing of Environment 129:122-131



13

Next Steps

The Reference Emission Level described in this report was developed following relevant guidance so that

it could be submitted to the UNFCCC and/or the FCPF as a proposed REL. However, there are a number of

items that should be addressed to improve the REL in a stepwise fashion:

A full ground-truthing effort should be conducted for the land cover and land cover change maps

Areas of active plantations must be identified and likely stratified out of the country’s forests

A forest inventory or forest carbon sampling plan should be developed and implemented, so that

country-specific emissions factors can be established.

A Monitoring, Reporting, and Verification System should be developed, consistent with the

methods used to develop the REL.

In the longer term, a reliable approach for estimating degradation should be chosen, and data

should be collected to implement such an approach.



14

1. INTRODUCTION

Reducing Emission from Deforestation and Forest Degradation, and enhancing forest carbon stocks

(REDD+) in developing countries is a valuable mechanism for countries that aim to mitigate the impacts of

climate change by cutting carbon dioxide emissions originating from the destruction of forests. In Liberia,

REDD+ is viewed as a potential opportunity and a viable source of sustainable finance for investment in

forest management, forest conservation, and forest restoration to enhance multiple benefits of REDD+,

including but not limited to biodiversity conservation, watershed management, enhanced resilient

capacity and poverty reduction9.

Liberia hosts a large percentage of the remaining forests within the Upper Guinean Forest Ecosystem that

stretches from Guinea to Togo. It covers an area of 111,369 square kilometres (43,000 sq. mi) and is home

to 4,503,000 people. Liberia possesses about forty percent of the remaining Upper Guinean rainforest.

The landscape is characterized by mostly flat to rolling plains with a thin strip of mangroves and swamps

along the coast and the plains rise to a rolling plateau and low mountains in the northeast. Liberia

experiences a conventional rainfall pattern and is kept wet for the most part of the year. The heaviest

rainfall occurs in June while the lightest rainfall is in December, with relative humidity between 90- 100%

during the rainy season and 60- 90% for the dry season. The geographical location of Liberia from the

equator makes the sun over head at noon throughout the year. The average temperature ranges between

28° C to 32°C in November and June respectfully.

Currently, around 85% of Liberians live below the international poverty line10 as they continue to recover,

economically and socially, from a decade of civil war that ended in 2003 and most recently from an Ebola

outbreak that took the lives of over 11,000 people. With a large percentage of the country forested, the

forestry sector has the potential to assist in the development of the country.

Deforestation and degradation drivers in Liberia include selective logging; pit-sawing; mining activities;

fuel wood and charcoal collection; the spread of shifting cultivation; permanent agriculture; and

anticipated increases in rubber and palm oil plantations11. Performance based climate financing may

provide Liberia with an alternative land use opportunity that will enable Liberia to financially benefit from

its forests without degrading them.

9 FCPF Readiness Assessment (2014): Mid-Term Report for Liberia. https://www.forestcarbonpartnership.org/liberia 10 Republic of Liberia Ministry of Internal Affairs. 2015. Overview of Liberia. Available at http://www.mia.gov.lr/2content.php?sub=210&related=40&third=210&pg=sp 11R-PP Country Submission for Liberia – 2012 http://www.forestcarbonpartnership.org/sites/fcp/files/2014/MArch/March/Liberia%20grant%20agreement.pdf

https://www.forestcarbonpartnership.org/liberia

http://www.mia.gov.lr/2content.php?sub=210&related=40&third=210&pg=sp



15

The Government of Liberia has already made strides in forest governance, in general, as well as particular

requisites for REDD+. Advancements include the creation and mobilization of necessary institutions and

frameworks with the mandate of supporting the development of REDD+ in the country, including the

National Climate Change Steering Committee (NCCSC), REDD+ Technical Working Group (RTWG) and

REDD+ Implementation Unit (RIU). Additionally, the Government has enacted laws and policies to

advance REDD+ strategy and submitted a Readiness Preparation Proposal (R-PP) to the Forest Carbon

Partnership Facility (FCPF), which was approved in March 2012. Liberia’s Voluntary Partnership Agreement

(VPA) for Forest Law Enforcement, Governance, and Trade with the European Union further supports

transparent and sustainable forest management. The historic partnership between the Governments of

Liberia and Norway announced in September 2014 further provides the financial and technical foundation

for successful implementation of REDD+ in Liberia. Liberia has also received assistance from the World

Bank’s Forest Carbon Partnership Facility (FCPF) to develop and apply strategies to reduce emissions from

deforestation and forest degradation.

Since the Readiness Preparation Proposal (R-PP) was signed in 2012, the RIU have partnered with several

organizations to complete components of the Readiness Package. This report serves as an input into this

effort through the development of a recommended REDD+ Reference Level. The report is structured with

the following sections:

Section 1: Explains Reference Levels and their technical components and provides further context of

REDD+ activities in Liberia.

Section 2: Provides an overview of the basic decisions required for the establishment of a reference

level, and identifies the most appropriate options for Liberia.

Section 3: Details the methods used to develop Liberia’s activity data, the extent of change in forests,

for deforestation.

Section 4: Details the methods used to estimate carbon stocks and develop emission factors for

Liberia.

Section 5: Combines activity data and emission factors to provide an estimate of historical emissions

and describes the projection of these historical emissions into the future to develop a REDD+

Reference Level.

Section 6: Identifies the next steps that should be taken to implement and improve the recommended

Reference Level.



16

Overview of REDD+ Reference Levels and MRV

In order for countries to receive credit for reducing net emissions under a REDD+ system, a benchmark

must be in place against which total emissions and removals are evaluated. Both the benchmark and

the evaluated emissions must be measured using internationally accepted methods.

The benchmark is referred to as a Reference Emission Level (REL) or Reference Level (RL). RELs represent

gross emissions from deforestation and forest degradation in a given time period, while RLs represent

both emissions and removals for all REDD+ activities. The REL/RL is based on historical greenhouse gas

emissions and removals, projected into the future under a Business as Usual (BAU) scenario in which no

actions are taken to reduce net emissions. Figure 1 provides a graphical depiction of historical emissions

used to develop an average reference level, which is compared to monitored emissions, the total

emissions and removals that are being evaluated against the reference level.

Figure 1. Hypothetical historical emissions (blue) used to develop an average reference level (red) that can be compared to

future monitored emissions (green).

REL/RL under the UNFCCC

This approach of measuring (MRV) against the benchmark (REL/RL) is established at the international

level. According to the Conference of the Parties (COP) to the United Nations Framework Convention

on Climate Change (UNFCCC) “forest reference emission levels and/or forest reference levels expressed

Em

issi

on

s, t

CO

2e/

yr

Year

Historical emissions Projected RL Monitored emissions

Initiation of REDD+ activities



17

in tons of carbon dioxide equivalent per year, are benchmarks for assessing each country’s performance

in implementing [REDD+] activities”.

The UNFCCC, in decision 4/CP.17 “invites parties to submit information and rationale on the

development of their forest reference emission levels and/or forest reference levels including details of

national circumstances.”

The UNFCCC calls for the development of a national reference level (RL) or reference emission level

(REL) for REDD+, based on decision 4/CP.15, which states:

“… in establishing forest reference emission levels and forest reference levels should do so

transparently taking into account historic data, and adjust for national circumstances.”

This was further elaborated in decision 4/CP.17 as follows:

“…modalities for the construction of forest reference levels and forest emission reference levels to

be flexible so as to accommodate national circumstances and capabilities, while pursuing

environmental integrity and avoiding perverse incentives.”

The development of RL/RELs can be seen as a two-step process. First is to produce estimates of

historical emissions and removals. These estimates are then projected into the future, and the

projections can potentially be adjusted, taking into account both national circumstances as well as

national capabilities.

The creation of forest RELs/RLs is guided by modalities contained in the UNFCCC COP decisions:

Use of historical data and adjustment for national circumstances should be transparent,

complete, consistent, and accurate

A step-wise approach is allowed to improve the forest RELs/RL by incorporating better data,

improved methodology, and additional pools where appropriate

Forest RELs/RLs are expressed in units of tons of CO2 equivalent per year ( t CO2e yr-1) and must

be consistent with the country’s GHG inventory

MRV

Once the RL/REL is established, it is compared against actual emissions, monitored under the

Monitoring, Reporting, and Verification (MRV) system. An MRV system must be in place to evaluate the



18

effects of REDD+ implementation in terms of emissions (green line in Figure 1) against reference levels

(red dotted line in Figure 1). In other words, the net emissions under REDD+ must be estimated within

an acceptable level of certainty in order to determine the difference between reference emissions and

actual emissions.

Internationally accepted methods to estimate REL/RL and MRV

The IPCC (Good Practice Guidance 2003, and Guidelines for National Greenhouse Gas Inventories

Agriculture, Forestry, and Other Land Uses [AFOLU] 2006) provides the framework for estimating

emissions and removals of CO2 in the AFOLU sector. The IPCC Guidelines refer to two basic inputs with

which to estimate greenhouse gas emissions and removals for the REL/RL and the MRV: activity data

and emissions factors.

Both historical emissions and emissions monitored under the MRV are estimated based on these inputs.

“Activity data” refer to the extent of an activity over a known time period. In the case of

deforestation, this is usually measured in terms of the change in areal extent of forest land,

presented in hectares over a known time frame (usually per year). This activity data can be

estimated separately for differing specific types of activities.

“Emission factors” refer to emissions/removals of greenhouse gases per unit of the activity

data. For deforestation, this would be per unit area, e.g. tonnes of carbon dioxide emitted per

hectare of deforestation.

“Activity data” combined with “emission factors” estimates the total amount of emissions/removals

taking place in a given year as a result of that activity. Emissions/removals resulting from land-use

conversion are manifested in changes in ecosystem carbon stocks, and for consistency with the IPCC

framework, we use units of carbon dioxide, specifically tonnes of carbon dioxide equivalence per

hectare (t CO2e ha-1), to express emission factors.

Three Approaches (Approaches 1-3) are presented as options in the IPCC guidance documents for

obtaining activity data, and three Tiers (Tiers 1-3) are presented as options for obtaining emission

factors (Table 1). Higher Approaches and Tiers correspond to greater detail in the underlying data,

whereas lower tiers rely extensively on generalized default factors. Different tiers can be used for

different components of RL/REL development, however for the most significant components it is

recommend that Liberia pursue a tier 2 or 3 in order to reduce uncertainty, while for less significant

components a tier 1 may be suitable.



19

For example, above ground tree biomass is by far the most significant carbon pool in forests, therefore

Liberia should seek tier 2 or 3, while litter and dead wood are not very significant carbon pools where

default factors could be the best option.

Table 1. IPCC Approaches and Tiers (note that it is not necessary to use the same level for activity data and emission factors)

Level Approach for activity data Tier for emission factor

1 Total area for each land use category,

but no information on conversions (only

net changes)

IPCC default factors

2 Tracking of conversions between land-

use categories

Country specific data for key

categories

3 Spatially explicit tracking of land-use

conversions

Detailed national inventory of

carbon stocks for key categories,

repeated measurements of

through time or modeling

While moving from Tier 1 to Tier 3 reduces the uncertainty range of GHG estimates, it also increases

the complexity and costs of measurement and monitoring. Likewise, achieving greater completeness

and certainty in a measurement and monitoring system means higher costs as it is likely that more

carbon pools would need to be monitored and that the monitoring would need to result in accurate

and precise estimates of emissions and removals.

1.1 Liberia’s Unique National Circumstances

Vast tropical forests cover nearly half of Liberia’s land mass, which are essential to the livelihoods of

Liberia’s peoples as well as the health of its ecosystems. While Liberia’s forests have historically been

subject to exploitation, compared to many of its neighbors, it has had relatively low deforestation rates.

In fact, according to some estimates, the country contains over half of West Africa’s remaining

rainforests12. Nevertheless, given the fact that over half of Liberia’s forest land has been allocated either

as commercial concessions or is designated for conservation as protected areas, the potential for

significant land use change and associated emissions in Liberia should be considered high. To date, most

12 http://www.euflegt.efi.int/liberia

http://www.euflegt.efi.int/liberia



20

of the concession areas have yet to be developed and protected areas are not yet well established, leaving

the future of Liberia’s forests unknown.

The country’s historically low deforestation rates can in part be attributed to political instability, which

slowed introduction of infrastructure and foreign investment that often accelerate land use change.

Liberia’s brutal civil war lasted over a decade and claimed 250,000 lives, and while the conflict was in large

part financed by natural resources including timber, the war prevented large-scale development and

agricultural expansion. During the country’s two civil wars, the economy collapsed, with the GDP declining

90% between 1987 and 1995 (Radelet, 2007). Although the economy saw a slight rebound after the first

civil war ended in 1996, it declined again during the second civil war. The second war ended in 2003, but

elections were not held until 2005. At that time, Liberia’s average income was one-quarter of the country’s

income in 1987, and one-sixth of that in 1979 (Radelet, 2007). After the 2005 elections, and the installation

of the new government, there was an acceleration in the pace of economic recovery.

As part of the country’s post-conflict development strategies, the Liberian government sought to develop

its forestry industry by offering tax incentives and issuing large timber concessions. However, corruption

and a lack of transparency resulted in unsustainable rates of timber harvesting and international scrutiny.

Following a reform process where significant efforts were made to regulate the logging industry and

establish a framework for sustainable forest management, a UN ban introduced to counter conflict timber

was lifted in 2006. Nevertheless, legal timber exports were slow to resume and commercial viability is

often challenged by the lack of roads and infrastructure to get timber to market13.

While historical rates of deforestation in Liberia have been relatively low, recent developments may be

altering this trend. Despite efforts made by national authorities to revoke illegally issued timber contracts

and control illegal exploitation of timber resources, loopholes in the system have been exploited by large

timber companies and enforcement of laws remains a key challenge in fostering a strong, well-regulated

timber industry. Furthermore, as Liberia’s post-war economy continues to stabilize, foreign investment

has grown, along with and expanded agricultural concessions for palm oil, rubber, mining, and timber. In

particular, there has been a significant rise in palm oil operations across the country through government

leases of land memorandums of understanding between international companies and local

communities14.

13 USAID Liberia. ‘Liberia – Environmental Threats and Opportunities’. 119/119 Assessment. 2014 14 http://www.theguardian.com/global-development/2015/jul/23/palm-oil-golden-veroleum-liberia-land-deals-ebola-crisis

http://www.theguardian.com/global-development/2015/jul/23/palm-oil-golden-veroleum-liberia-land-deals-ebola-crisis

http://www.theguardian.com/global-development/2015/jul/23/palm-oil-golden-veroleum-liberia-land-deals-ebola-crisis



21

The deadly Ebola outbreak in Liberia primarily between 2013 and 2014 also likely had a significant impact

on deforestation rates in Liberia. The scale of the outbreak, claiming thousands of lives across the country,

brought the national economy, foreign investment, and infrastructure initiatives to a standstill. The World

Bank estimated that the fiscal impact to Liberia was a loss of $93 million, reducing GDP by 3.4 percent.15

In response to this economic impact and foreseen loss of development momentum generated over the

past decade, a large amount of funding is being pledged to countries hardest hit by the disease. As of

April 2016, the World Bank alone mobilized $385 million for Liberia’s recovery effort, which in addition to

facilitating the rebuilding of the country’s public health system, includes financing for economic recovery

and commercial financing to enable trade, investment and employment.16 While these investments will

likely contribute to the improved livelihoods of Liberia’s crisis-afflicted populations, it may ultimately

result in increased rates of deforestation associated with development of the country’s economic

resources in years to come.

As Liberia strives to stabilize in the wake of the Ebola outbreak, current drivers of deforestation and forest

degradation are likely to persist and intensify. In particular, deforestation and degradation activity is likely

to increase among forest concessions and palm oil concessions. Forest concessions comprise the largest

official category of land use by area in Liberia, and 29% of these areas are located in dense, carbon-rich

forests (LTS International 2016)

Palm oil dominates industrial agriculture land use, and current concessions make up 5% of the total forest

area. These concessions lands are overwhelmingly operated by a small set of large multinational

companies (Sime Darby, Golden Veroleum (GVL), Equatorial Palm Oil (EPO), and Maryland Oil Palm

Plantations (MOPP)), and according to the LTS International Forest Cover and Land Use Analysis Draft

Report (2016), this serves as the basis for the scale and location of the industry for the next century. The

same report stated that the area of land cleared for oil palm plantation in the next 10-15 years is estimated

to reach 250,000 ha based on current industry plans and the pace and scale of development among these

concession areas is predicted to accelerate to ensure profitability of these large land acquisitions.

In addition, mining activity is likely to intensify as interests seek to exploit the country’s rich mineral

resources, as was also highlighted in Liberia’s R-PP. Mineral exploitation licences have been granted over

4.6 million hectares of land and the extraction of iron ore in Liberia is accelerating. At least six iron ore

concession agreements have been signed with a total estimated investment value of $13 billion, and the

future area impacted by mining is estimated to be between 137,200 and 200,800 hectares (LTS

15 https://csis-prod.s3.amazonaws.com/s3fs-public/legacy_files/files/publication/Runde_Savoy.pdf 16 http://www.worldbank.org/en/topic/health/brief/world-bank-group-ebola-fact-sheet

https://csis-prod.s3.amazonaws.com/s3fs-public/legacy_files/files/publication/Runde_Savoy.pdf

http://www.worldbank.org/en/topic/health/brief/world-bank-group-ebola-fact-sheet



22

International 2016). This figure accounts only for the formal mining sector, and not artisanal mining, which

is common in Liberia and may have a significant cumulative impact.

All of these factors – war, the Ebola outbreak, and recent and projected economic development – indicate

that the past circumstances of Liberia’s economy and land use do not serve as representative indicators

of its future.

REDD+ Efforts in Liberia

Since the Readiness Preparation Proposal (R-PP) was signed in 2012, the RIU have partnered with several organizations to complete components of the Readiness Package (R-Package), including the Development of Reference Scenario for REDD+ Readiness. Below are summaries of other components of the R-package, MRV development efforts, and the existing REDD+ pilot project in Liberia.

Land cover mapping – Metria/Geoville

In February 2014, the Forestry Development Authority signed a contract JV Metria/GeoVille to conduct

a comprehensive land cover and forest mapping in Liberia. The Land Cover and Forest map is based on

satellite imagery from Landsat 8 and RapidEye. The integrated mapping results are now prepared for

delivery to FDA in digital formats as well as printed maps. JV Metria/GeoVille has now completed the

final phase of integrating the mapping results of Liberia´s Land Cover and Mapping performed under

contract from the Forestry Development Authority.

Liberia-National REDD+ Strategy Consultation

In July 2014, the Forestry Development Authority signed a contract with LTS International and NIRAS to

develop the National REDD+ Strategy. The objective is to develop an integrated national REDD+ strategy

through a participatory and transparent consultative process with REDD+ stakeholders. The REDD+

Strategy will be prepared in conjunction with a Strategic Environmental and Social Assessment (SESA).

The key output of this assignment is to provide analysis on Land Use Options, REDD+ Strategy Options

and the Policy, Legal and Institutional Framework, national REDD+ strategy, REDD+ road map and action

plan. A draft REDD+ strategy options report was submitted for stakeholder review, and a final draft has

been submitted to the RIU.

Strategic Environmental and Social Assessment (SESA)

In May 2014, the Liberian Forestry Development Authority (FDA) signed a contract with Tetra Tech ARD

assist in the preparation of a Strategic Environmental and Social Assessment (SESA) and a draft REDD+

Environmental and Social Management Framework (ESMF). The technical oversight and coordination

of the SESA and ESMF is provided by the Environmental Protection Agency (EPA) of Liberia, in

accordance with the environmental law, through a SESA coordinator and a stakeholder SESA working



23

group. The SESA is to contribute to the REDD+ Readiness process in Liberia by assessing how REDD+

strategy options address environmental and social priorities associated with current patterns of land

use and forest management. Gaps identified through this assessment would lead to adjustments in the

REDD+ strategy options to close the gaps. Also, the SESA will provide an Environmental and Social

Management Framework (ESMF) that will outline the procedures to be followed for managing potential

environmental and social impacts of specific policies, actions and projects during the implementation

of the REDD+ strategy that is finally selected. The SESA Inception Phase and Report were completed in

September 2014, and the team is in the process of completing the remaining deliverables.

REDD+ Communication Strategy and Information Sharing

In March 2014, the Liberian Forestry Development Authority (FDA) signed with Fauna and Flora

International to develop the REDD+ Communication Strategy and Information sharing to contribute to

the successful implementation of the REDD+ Strategy in Liberia. The objectives of this assignment are

to conduct a communication analysis for the REDD+ process in Liberia as envisaged in the R-PP, design

a comprehensive and coherent REDD+ Communication strategy that will enable the RIU to accomplish

the following:

a. Design a Communication and Information Sharing Strategy targeted at Key Stakeholders Groups and their constituencies

b. Prepare and produce appropriate local language and accessible media for this strategy, including best practice key messages

c. Conduct media campaign to promote REDD+ awareness at National, County and Stakeholder levels through newspaper, radio and TV

d. Raise the public profile of the National REDD+ Programme locally, nationally, regionally and with all identified audiences;

e. Ensure effective lobbying and advocacy with critical stakeholders for buy into the REDD+ dialogue and implementation;

f. Employ an effective communication approach useful for excellent expectation management.

Feedback and Grievance Redress Mechanism

In April 2016, the feedback and grievance redress mechanism contract was awarded to PARLEY Inc. to

establish a Grievance Mechanism to address risks of dispute or conflict between stakeholders that may

arise as a result of the Readiness preparation challenges. These may include issues relating to

commitments made by the project, land, benefit sharing, community rights. The intention of the

grievance redress mechanism (GRM), as part of the governance arrangements for the REDD+ Project, is

to promote effective channels for citizen feedback and redress so as to improve the credibility and

performance of the overall program. The products of this contract are expected to be completed June

2016.



24

MRV efforts and Norway-Liberia Partnership

In October 2015, the Wageningen University facilitated consultation workshop on the development of

capacities for a National Forest Monitoring and Measurement, Reporting and Verification (MRV) System

to support REDD+ participation of Liberia. A draft of the MRV roadmap term of references document

was reviewed by stakeholders through the REDD+ Implementation Unit and REDD+ Technical Working

Group, and the final version was produced on June 7, 2016. The final MRV roadmap recommends key

next steps for Liberia to improve its REDD+ NFMS/MRV capacities. It is important to note that some of

the recommended steps have already been taken or are currently underway.

Wonegizi REDD+ Pilot Project

The Wonegizi REDD+ Pilot project, developed by Fauna & Flora International (FFI) together with FDA and

local NGO Skills & Agricultural Development Services (SADS), aims to lower greenhouse gas emissions from

deforestation by reducing agricultural pressures on the Wonegizi forest (and Proposed Protected Area)

in Lofa Country, northwest Liberia (approximately 37,968 ha). By introducing community management

of the protected area and offering technical support and funding to increase the efficiency of land use

and agricultural practices, project proponents intend to lower the use of slash-and-burn agriculture by

local populations.

The project was designed for validation through the Plan Vivo Standard, along with biodiversity components

compatible for validation under the Climate, Communities, and Biodiversity (CCB) Standard Gold Level

certification requirements. Plan Vivo accepted the project idea note (PIN) in January 2014, and in 2015 pre-

validation was undertaken by an external auditor. However, due to the size of Wonegizi, and to allow for even

greater expansion at the landscape scale, the Wonegizi REDD+ Pilot project will now seek duel certification under

the VCS and CCB. The project was initiated in 2012 and is expected to continue for an initial period of

over 10 years, with project activities expected to reduce deforestation by an estimated 55% and forest

degradation by 60%. This corresponds to total projected emission reductions of around 354,158 t CO2-

e over 10 years and 797,013 t CO2-e over 50 years (FFI & RSS GmbH, 2014), although compliance with

VCS requirements might result in somewhat lower estimates.

1.2 A Note on Available Data

Development of a REDD+ Reference Level and a functioning REDD+ program in general requires significant

amounts of data on forest cover, forest use, forest inventory, infrastructure, development plans, and

economics. While some data is available for Liberia, there is a lack of complete country-specific

information, and in some cases there are multiple sources of data that are inconsistent with no indication

of which source provides the most accurate information. This is not unusual as a country begins the

process of developing a REDD+ program, however, it is critical that a system for updating, maintaining,



25

and storing data is developed and followed, to allow Liberia to ensure accurate accounting and

transparency, and to maximize the efficiency of the REDD+ program.

For this report, data were gathered from numerous sources, including FDA, LISGIS, the University of

Maryland, literature searches, global datasets, and partners working on other components of the R-

Package. A limited field campaign was also conducted in early 2016. Every effort was made to ensure that

the most accurate and current data were used, and data sources are noted throughout the report.

2. DEFINING THE SCOPE OF THE REL/RL

2.1 International Guidance on RL Development

There are two main sources of guidance on the development of a REDD+ Reference Level, the United

Nations Framework Convention on Climate Change (UNFCCC) and the World Bank Carbon Fund. The

UNFCCC provides general recommendations for the development of an internationally acceptable

Reference Level, while the Carbon Fund has a Methodological Framework that includes stricter

requirements that must be met in order to receive funding. All of these systems refer to accounting

methods described by the Intergovernmental Panel on Climate Change (IPCC).

UNFCCC Conference of Parties (COP) decisions contain modalities that guide the development of forest

reference levels, particularly decision 12/CP.17 and its Annex. According to these modalities, Parties must

be transparent in establishing RLs, taking into account historical data and, if appropriate, adjusting for

national circumstances17. While forest RLs can be developed sub-nationally as an interim measure while

transitioning to a national scale, Liberia has chosen to develop its RL at a national scale. A step-wise

approach may be used, allowing Parties to improve the forest Reference Level (REL) by incorporating

better data, improved methodologies and additional pools, if appropriate. Forest RLs are expressed in

units of tons of CO2 equivalent per year and must maintain consistency with a country’s greenhouse gas

inventory (according to 12/CP.17, Paragraph 8). In response to the guidelines for submissions of

information on Reference Levels provided in decision 12/CP.17, a summary of Liberia’s decisions on these

modalities is given in Table 2. Further descriptions on each of these modalities is described in the

remainder of this section.

17 Decision 4/CP.15, paragraph 7.



26

Table 2. UNFCCC modalities relevant for Liberia's national Reference Level

Reference to Guideline

Description Liberia’s Proposal

Decision 12/CP.17 Paragraph 10

Allows for a step-wise approach

REL is at national scale, and includes all drivers of deforestation

Degradation will be added as additional data become available.

Decision 12/CP.17 Annex, paragraph (c)

Pools and gases included Pools: (activity specific) - Aboveground and belowground biomass - Dead wood - Litter - Soil carbon

Gases: - Include CO2

Decision 12/CP.17 Annex, paragraph (c)

Activities included Include deforestation caused by agriculture, mining, forestry infrastructure, and other infrastructure

Other activities will be included in step-wise improvements of the RL

Decision 12/CP.17 Annex, paragraph (d)

Definition of forest used is same as that used in national GHG inventory

Minimum tree cover: 30%

Minimum height: 5 m

Minimum area: 1 ha

Decision 12/CP.17 Annex

The information should be guided by the most recent IPCC guidance and guidelines,

All data are gathered using best practices and integrated to estimate emissions using IPCC 2003 and 2006 guidelines18

Where country specific data are not available, they will be developed

Decision 12/CP.17 II. Paragraph 9

To submit information and rationale on the development of forest RLs/RELs, including details of national circumstances and on how the national circumstances were considered

Liberia proposes to make adjustments to allow for national circumstances because historical emissions are likely not good indicators of future emissions.

18 The two IPCC reports used are the IPCC 2003 Good Practice Guidance for the LULUCF sector (IPCC 2003 GPG) and the IPCC 2006 Guidelines for National GHG Inventories, Volume 4 AFOLU (IPCC 2006 AFOLU)



27

The World Bank Forest Carbon Partnership Facility Carbon Fund was designed to provide incentives for

countries to pilot implementation of REDD+ programs. To receive funding under the Carbon Fund,

countries must adhere to the guidelines described in the Methodological Framework (FCPF 2013). There

are five basic considerations that need to be addressed in the establishment of historical emissions to

develop a REDD+ reference level that will be discussed in this section:

Finalize a forest definition

Determine the scope of activities

Establish the reference period

Determine the scale

Identify the pools and gases to include

A summary of each of these considerations, and how they are addressed by both the UNFCCC and the

Carbon Fund is provided below, along with recommended actions. The potential adjustment of historical

emissions based on national circumstances is discussed in Section 5.

2.2 Forest Definition

For the purposes of REDD, forest is defined in terms of minimum thresholds for canopy cover, height and

area. According to the FAO and various UNFCCC decisions, including the Marrakech Accords (UNFCCC

2001), forest is defined on a country basis, with a minimum area of land between 0.05 and 1 hectares,

with minimum tree canopy cover of 10-30%, and the potential to reach a minimum height of 2-5 m at

maturity in situ.

In late January 2016, FDA sponsored a workshop on Liberia’s Forest Definition. The workshop was held in

Lofa County, and was attended by a broad cross section of stakeholders, from the government, civil

society, and international NGOs. During the 5 day workshop, the options for and implications of Liberia’s

forest definition were discussed. At the completion of the workshop, a final forest definition was chosen,

with the following thresholds:

Minimum area of one hectare

Minimum canopy cover of 30 %

Minimum height at maturity of 5 meters

It was further decided that agricultural plantations, including tree crops such as palm, rubber and cacao,

would not be considered forest under Liberia’s definition.



28

2.3 Scope

Often, forest reference levels and reference emission levels are considered one and the same. However,

some consider reference levels to include both emissions and removals of greenhouse gases, while

reference emission levels address only emissions. Different entities have different specifications for what

must be included in a reference level.

The United Nations Framework Convention on Climate Change 19 encourages undertaking activities,

including reducing emissions from forest degradation, as deemed appropriate and in accordance with

existing capabilities and national circumstances.

The Forest Carbon Partnership Facility (FCPF) Methodological Framework 20 states that Emission

Reduction (ER) Programs can choose which REDD+ activities and sources and sinks to include in the ER

Program Reference Level. ER Programs are required to account for emissions from deforestation at a

minimum, and emissions from forest degradation should be included where they are significant:

“Emissions from forest degradation are accounted for where such emissions are more than 10% of

total forest-related emissions in the Accounting Area, during the Reference Period and during the

Term of the emission reduction purchase agreement (ERPA). These emissions are estimated using

the best available data (including proxy activities or data)”

In general, deforestation must always be addressed in a REDD+ system, and forest degradation activities

should be included when at least one of the following conditions exist:

A specific forest degradation activity results in significant emissions,

Capacity and resources exist to reliably measure and monitor those emissions cost-effectively,

There is potential that interventions could reduce such emissions.

While it is possible to obtain a reasonable initial estimate of deforestation from global datasets, it is much

more difficult to achieve an accurate picture of degradation. In assessing data currently available for

Liberia, it is clear that accurate data relevant to degradation are very limited, at a country, regional, or

even global level. This makes it very difficult to estimate emissions from degradation with any certainty;

data that are available provide estimates of degradation ranging from 8% to nearly 50% of total emissions

19 UNFCCC 1/CP.16 Paragraph 70: http://unfccc.int/resource/docs/2010/cop16/eng/07a01.pdf 20 FCPF Carbon Fund Methodological Framework, December 20, 2013, Criterion 3: https://www.forestcarbonpartnership.org/carbon-fund-methodological-framework

http://unfccc.int/resource/docs/2010/cop16/eng/07a01.pdf

https://www.forestcarbonpartnership.org/carbon-fund-methodological-framework



29

from land use and land use change, depending on the methods used21 . In order to improve these

estimates, it is necessary for Liberia to undertake substantial effort to acquire appropriate data.

Because it is not possible with the current data available to develop reliable estimates of emissions from

degradation, it is recommended that Liberia focus at present on assessing emissions from deforestation.

As capacity increases over time, emissions from degrading activities, as well as removals from

enhancements, can be incorporated in a step-wise approach. We therefore recommend that Liberia use

a Reference Emission Level at present, focusing only on emissions. Moving forward, the country should

work towards including removals and developing a Reference Level. Annex 1 provides additional detail

on estimates of emissions from degradation and recommendations for improving these estimates in the

future, so that degradation can be included in the Reference Level and the REDD+ program.

2.4 Reference Period

The historical reference period is the period from which data on past changes in forest area are

established, analyzed, and projected into the future. It is used to determine the average annual level of

emissions against which future years are compared. There are a number of factors that must be

considered in determining an appropriate reference period. This period, therefore, is dictated in part by

available data. In the case of Liberia, there are reliable data available on forest loss from 2000 through

2014. The Carbon Fund Methodological Framework (Revised Final, June 22, 2016) states that the end year

of the reference period should be “the most recent date prior to two years before the TAP starts the

independent assessment of the draft ER Program Document and for which forest cover data is available

to enable IPCC Approach 3 (Indicator 11.1). Additionally, the start date for the reference period must be

about 10 years before the end date, unless an exception is requested and granted, in which case it cannot

be more than 15 years before the end date (Indicator 11.2).

Given Liberia’s unique circumstances (described in detail in section 1.1), with the second civil war ending

in 2003, timber sanctions enacted between 2003 and 2006, and economic decline through 2005, land use

and land cover change have increased in recent years, as the country’s economy begins to improve. In

fact, the average annual rate of forest loss between 2002 and 2006 is 0.19%, while the average annual

rate between 2009 and 2013 is 0.61%. The development of annual land cover maps created within this

project found a spike of deforestation in 2013. This is likely due to an increase in land use activities, but

also may be due to the fact that in 2013, a new Landsat satellite was launched, Landsat 8, which has

improved image quality and observation frequency relative to its predecessors. This satellite therefore

likely improved detection of deforestation that occurred at some point in the recent past, especially given

21 Additional information provided in Annex 1.



30

considerably less Landsat data were collected in 2012 than in previous years following the failure of one

of the two satellites collecting data in tandem since 2000, Landsat 5, in November 2011.

Section 3.1 presents the annual deforestation and resulting deforestation rate, and based on this analysis

and the recent history of Liberia, a reference period of 2005-2014 is recommended. This period is in line

with the revised FCPF Methodological Framework.

It should be noted that UNFCCC guidance on reference period is far less prescriptive than the FCPF, and

allows for more flexibility.

2.5 Scale

To ensure consistency and a unified approach from the inception of the REDD program, Liberia’s

Reference Emission Level has been developed at the national scale. Such an REL can be applied at the

district level as needed. The advantage of a national approach is that the integration of separate

subnational RELs and MRV systems is not necessary. Therefore, the process of developing an REL is

simplified and can happen more quickly than if common standards and agreements had to be developed

for subnational jurisdictions to use. However, there are existing efforts towards REDD, notably the

Wonegizi Community REDD+ Pilot, which is currently in its fourth year. To allow for appropriate accounting

of emission reductions as well as equitable benefit distribution, it is recommended that Liberia adopt a

nested approach for REDD+ implementation, ensuring that the efforts of existing and future projects are

encouraged by the national REDD+ program.

There are varied ways that nesting can be undertaken, and a number of issues that must be considered to

guarantee that there is alignment between project and national accounting. Primary among these are the

activities included and the methods used for establishing reference levels/baseline and monitoring

performance. If there are incongruities between project and national systems or accounting, it is necessary

to take steps to rectify the incongruities. As the existing project at Wonegizi undergoes final verification,

it will be necessary to assess how it relates to the national reference level and REDD+ program and work

to align the two. Guidance on implementing nested approaches to REDD and addressing technical

considerations of nesting are available through the USAID LEAF Planning Guide – Integrating REDD+

Accounting within a Nested Approach22 and the VCS Guidance Document: Options for Nesting REDD+

Projects23.

22 Available at http://www.leafasia.org/library/planning-guide-integrating-redd-accounting-within-nested-approach. 23 Available at http://www.v-c-s.org/wp-content/uploads/2016/07/Nesting-Options-1-Jul_Eng_final.pdf.

http://www.leafasia.org/library/planning-guide-integrating-redd-accounting-within-nested-approach

http://www.leafasia.org/library/planning-guide-integrating-redd-accounting-within-nested-approach

http://www.v-c-s.org/wp-content/uploads/2016/07/Nesting-Options-1-Jul_Eng_final.pdf



31

2.6 Pools and gases

The most significant carbon pool in forests is typically aboveground live tree biomass, and this should be

included in all accounting for Liberia’s REL and MRV. In addition, belowground live tree biomass can easily

be estimated using a root to shoot ratio. Dead wood and litter usually contribute a much less significant

percentage of total forest carbon, and often require a significant investment of resources to measure with

accuracy. However, it is easy to estimate the extent of these pools using IPCC default values, and this

should be done until and unless it is determined that these pools are significant enough that they warrant

actual measurements. Change in soil organic carbon can be estimated and included based on IPCC

methods. The following are the included pools for Liberia’s deforestation REL:

Aboveground live tree biomass (using available datasets, and inventory data when available)

Belowground live tree biomass (based on root:shoot ratio)

Dead Wood (IPCC default ratio)

Litter (IPCC default ratio)

Change in Soil Organic Carbon (based on soil carbon from World Harmonized Soil Databased

and IPCC default factors for change)

When degradation is added to Liberia’s Reference Level, harvested wood products should likely be

included, especially in the case of degradation from timber harvesting.

The primary gas that is emitted from land use change is carbon dioxide, and this is the only GHG included

in the RL.

3. ACTIVITY DATA

As described in Box 1 above, the estimation of emissions will be based on a combination of activity and

emission factors for the various activities. In addition, Section 2.3 describes the recommendation that

Liberia focus its reference level development on Deforestation with a Reference Emissions Level.

Therefore, historical Activity Data for deforestation have been calculated for the recommended Reference

Period. This information is then used to estimates historical average rates of deforestation over the

Reference Period. For degradation, at this point the lack of existing information prevents a detailed or

accurate estimate of degradation activity data to be completed. Therefore, instead, two approaches for

creating an initial estimate of degradation are presented in Annex 1, to better understand the potential

magnitude of degradation emissions, especially in comparison to deforestation emissions.



32

To estimate land use and forest cover change from deforestation, existing spatial data sets were first

compiled and evaluated for their applicability, completeness, and accuracy. Based on the assessment, the

most applicable spatial data were used to create land cover change products for 2000-2014. These maps

were used to create annual estimates of the area of change for different land cover classes over this time

period along with the average annual land cover change for the recommended Reference Period 2005-

2014. The implications of various stratification options were also examined. The detailed methods to

create these products are presented here along with associated Appendices.

3.1 Evaluation of existing spatial datasets

To evaluate deforestation, the existing data sets were examined and assessed for applicability in

estimating current and historical land cover (Appendix A). Based on this analysis, it was determined that

the recently produced 2014 Land Cover map for Liberia (Metria/GeoVille 2016) would be used as the basis

for nonforest and forest stratification. An approach was then developed and applied to estimate historical

deforestation rates for each of the forest strata defined by the 2014 land cover map and the 2000-2014

forest loss product developed by Hansen et al (2013). This approach is described in detail below.

2014/15 forest stratification dataset

The high resolution (10m) land cover map for the whole of Liberia for 2014 that has been developed based

on RapidEye and Landsat 8 data splits the landscape into 10 different land cover classes (Metria/GeoVille

2016):

Tree cover >80 %

Tree cover 30-80 %

Tree cover <30 %

Mangrove and swamp

Settlements (urban and rural)

Surface water bodies

Grassland

Shrub

Bare soil

Ecosystem Complex (rock and sand)

According to the recently adopted forest definition, two of these classes are considered forest in Liberia:

tree cover >80% and tree cover 30-80%. This map serves as the best and most recent forest classification

for Liberia, and will therefore be used as the minimum level of stratification of forest cover in the

Reference Level. The Metria/GeoVille product uses a minimum mapping unit of 0.5 hectares; however,



33

the minimum area in the recently adopted forest definition is 1 hectare. Therefore, for the purposes of

estimating area of forest loss, all forest areas less than one hectare were excluded from the analysis of

deforestation. Pixels in these <1ha forest patches were reclassified in the ‘tree cover <30 % canopy cover’

class. Removing patches of pixels with >30% canopy cover but covering less than 1 ha reduced the total

estimated area of forest in Liberia in by only 0.75 % compared to the original Metria/Geoville map (see

Table 3 below). Using these criteria, the total forest area then can be calculated to comprise 68% of

Liberia’s land cover.

Table 3. Area of each forest class in 2014/15 with a 0.5 ha and 1 ha minimum forest area size. (Note that at the same time as

this analysis was performed, the data were transformed from the 10m UTM projection of the M/G map to a 0.00025 degree

(~30m resolution) latitude/longitude projection to match the Hansen et al. (2013) layers used later. This results in the slight

difference in the total area. All area calculations are however performed in a UTM projection throughout.)

Forest class Original 2014/15 Metria/Geoville Map with 0.5 ha minimum forest threshold and 10 m pixels size (ha)

Processed Metria/Geoville 2014/15 Map with a 1 ha minimum forest threshold and 30m pixel size (ha)

Forest <80% canopy cover

4,389,270 4,375,862

Forest 30-80 % canopy cover

2,186,495

2,150,657

1-30 % canopy cover

1,529,949

1,579,147

Classes without trees

1,467,152

1,467,424

Total 9,572,866 9,573,090

As stated above, in the recently finalized forest definition, plantations are excluded from the forest

definition of Liberia, even if land cover and trees contained would otherwise meet the condition for forest.

Thus for this forest definition to be properly reflected in the land cover maps, plantations should be

included as a separate non-forest strata, however using current datasets this cannot be accurately

performed. Accurate spatial layers showing exactly where oil palm and rubber have been planted (as

opposed to the broad concession boundaries) are not currently available. As few large-scale plantations

existed in the early 2000’s this is not as large an issue for the start of the reference period maps. However,

the Metria/Geoville map currently does not distinguish planation areas and thus, some areas in its >30%

canopy cover classes are in fact plantations, and are therefore not included in Liberia’s definition of forest.

An additional effort is required to map all actual plantation areas (as opposed to plantation concession

boundaries). Since plantation areas are difficult to map using remote sensing and automated methods,



34

and instead require considerable local experience or data, it is recommended that this effort involve on-

the-ground teams and collaboration from plantation companies.

An independent field effort was undertaken in early 2016, to assess the accuracy of the percent forest

cover cut-offs of the forest classes of the Metria/Geoville product. The purpose of this exercise was to

check that there were no biases that could adversely influence its use in the historical deforestation

analysis based on Hansen et al (2013) data. This limited sampling found the Metria/Geoville map to have

an accuracy of 92.6%, with no apparent systematic bias between classes and no evidence that the 80%

canopy cover cut-off was incorrectly applied. The methods and analysis for this ground-truthing effort are

described in Appendix B. This also allowed for the creation of 95% Confidence Intervals around the

landcover class areas shown in Table 3, themselves with considerable uncertainty.

2005 forest stratification dataset development

In order to estimate land cover change within each forest stratum over time it is necessary to know the

land cover at the start of the reference period. While there are earlier land cover maps for Liberia, they

are all at a scale that is not appropriate for estimating land cover change with any accuracy (Appendix A).

For instance, a very coarse product was developed in 2004 by Bayol and Chevalier, with a minimum

mapping unit of 10 km2. In addition, no country-wide spatially-explicit estimate of

deforestation/degradation has been performed for Liberia post-2000 (See Christie et al. 2007 for an

analysis of land cover change from 1986-2000). Details of available land cover and land cover change data

and their relevance are provided in Appendix A.

In 2013, Hansen et al published a set of global spatial products that span 2000 to 2013. This includes a

global 30-meter resolution ‘2000 Percent Forest Canopy Cover’ map and a 30-meter resolution ‘Annual

Forest Loss’ product produced annually from 2000 onwards. The Hansen et al (2013) ‘2000 Percent Forest

Canopy Cover’ precisely matches the ‘Annual Forest Loss’ product produced from 2000 onwards (Hansen

et al 2013), which can be used to create a series of land cover change products and thus create annual

deforestation activity data. The Annual Forest Loss product analyzes all available Landsat imagery and

combines training data from across the planet to estimate forest loss from (currently) 2000 to 2014

(Hansen et al 2013).

These data have been freely released by the University of Maryland24 and will be annually updated. The

datasets have been widely publicized through Global Forest Watch25 and the use of such datasets is

24 http://www.earthenginepartners.appspot.com/science-2013-global-forest 25 http://www.globalforestwatch.org/



35

recommended in the Global Forest Observations Initiative’s Methods and Guidance Document

“Integrating Remote-sensing and Ground-based Observations for Estimation of Emissions and Removals

of Greenhouse Gases in Forests”26 (referred to as ‘MGD’). GFOI MGD explicitly encourages the use of this

dataset and now has a module advising on how this dataset can be used in the development of country-

level reference levels27. However, the MGD and various research indicates that the Hansen et al (2013)

can have significant local biases in its canopy cover estimates, and thus need local correction before being

used for Liberia.

Based on this evaluation, it was concluded that the Hansen et al dataset provides the highest resolution

data available to identify forest loss on an annual basis. This dataset was therefore used to produce annual

forest strata maps and annual deforestation estimates for the reference period, 2005-2014. These were

based on the Hansen 2000 canopy cover map, corrected for Liberia using the 2014 Metria Geoville

classifications, with annual forest loss calculated from that date onwards. As recommended by the GFOI

MGD, the data was first processed to create a ‘Liberia-Corrected 2000 Percent Canopy Cover’ map,

stratified by forest class (30-80% Canopy Cover; >80% Canopy Cover), using a combination of information

from the Metria/Geoville map and the published Hansen et al (2013). The steps to accomplish these tasks

are explained in Appendix C.

3.2 Deforestation Rate Estimation

Using the approach described in Appendix C, Hansen et al’s (2013) Annual Forest Loss product was also

Liberia-corrected and stratified by forest canopy cover class to produce an annual land cover change

product from 2000-2014 (Figure 2). From these datasets, the area of forest in each forest class at the start

of the Reference Period, the year 2005, was calculated along with subsequent annual area of forest loss.

Table 4 shows the annual forest loss for the two forest strata and Table 5 shows the annual rate of forest

loss by forest strata. Over the Reference Period, the annual deforestation rate was 0.46%28. For forests

>80% canopy cover, the annual deforestation rate was 0.36%, while for forests with 30-80% canopy cover,

the annual deforestation rate was 1.07%.

26 http://www.gfoi.org/methods-guidance/ 27 http://www.gfoi.org/wp-content/uploads/2015/03/MGDModule2_Use-of-Global-Data-Sets.pdf 28 For comparison, the annual deforestation rate on the REDD+ pilot project in Wonegizi was estimated at 0.24%, according to the Technical Specification (FFI & RSS GmbH, 2014). This rate was derived from a simple historical analysis of forest cover change within the PPA boundaries between the period of 2001 to 2013.




36

Figure 2. Forest cover and change from 2002-2013. Note that Reference Period is 2005-2014; additional years included for

context. (Based on Liberia-corrected 2000 Percent Forest Cover, Forest Cover Change product (Hansen et al 2013) and 2014

Liberia Land Cover Map (Geoville/Metria 2016)



37

Table 4. Annual forest loss in hectares since 2001 and over reference period 2005-2014, by forest cover class

Forest >80% Forest 30-80% All forest

Total initial area in 2000

Forest area (ha)

5,737,119 926,047 6,663,166

Forest Loss (ha)

2001 9,635 3,310 12,945

2002 17,848 7,579 25,428

2003 6,675 3,079 9,754

2004 2,407 1,196 3,603

2005 3,559 1,845 5,403

2006 10,735 6,597 17,331

2007 14,742 8,361 23,103

2008 11,100 7,288 18,389

2009 28,528 15,170 43,698

2010 8,764 3,630 12,393

2011 13,428 6,039 19,467

2012 27,093 11,530 38,623

2013 47,692 21,100 68,792

2014 39,496 15,912 55,408

Total forest loss over reference Period

205,135 97,472 302,608

Average annual loss over reference period

20,513.5 9,747 30,261



38

Table 5. Annual rate of forest loss since 2001 and over reference period 2005-2014, by forest cover class

Year deforested Forest >80% Forest 30-80% Combined forest loss

percent loss

2001 0.17% 0.36% 0.19%

2002 0.31% 0.82% 0.38%

2003 0.12% 0.34% 0.15%

2004 0.04% 0.13% 0.05%

2005 0.06% 0.20% 0.08%

2006 0.19% 0.73% 0.26%

2007 0.26% 0.93% 0.35%

2008 0.20% 0.82% 0.28%

2009 0.50% 1.71% 0.67%

2010 0.16% 0.42% 0.19%

2011 0.24% 0.70% 0.30%

2012 0.48% 1.34% 0.60%

2013 0.85% 2.48% 1.07%

2014 0.71% 1.92% 0.87%

Average annual loss over reference period

0.36% 1.07% 0.46%

Figures 3a shows the annual forest loss in hectares from 2000-2014 for both forest strata. Figure 3b shows

the cumulative forest loss since 2000 for forests with >80% canopy cover, and Figure 3c shows cumulative

forest loss since 2000 for forests with 30-80% canopy cover. Because there is more total forest with >80%

canopy cover, there was more total deforestation in this stratum, while the annual rate of deforestation

was much lower.



39

Figure 3a. Annual Forest Loss over time (ha); Purple lines indicate extent of ‘Reference Period’

Figure 3b. Cumulative Forest Loss over time for tree cover >80% (ha); Purple lines indicate extent of ‘Reference Period’

0

10,000

20,000

30,000

40,000

50,000

60,000

2000 2002 2004 2006 2008 2010 2012 2014

Are

a o

f A

nn

ual

Fo

res

Lo

ss (

ha)

Forest >80%

Forest 30-80%

0

50,000

100,000

150,000

200,000

250,000

300,000

2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014

Hec

tare

s

Forest >80%



40

Figure 3c. Cumulative Forest Loss over time for tree cover 30-80% (ha); Purple lines indicate extent of ‘Reference Period’

3.3 Characterization of Deforestation Trends

In general deforestation appears to be of two types, with large clearings associated with the creation of

plantations, visible as five clusters each 20-50 km from the coast stretching down the whole coastline, and

a network of tiny clearings located broadly throughout the country though focused around roads and

settlements (Figure 2). The small clearings are likely associated with agriculture and logging, and are

relatively evenly spread throughout the time period, whereas the plantation clearings initiate from 2011

onwards.

Additional analyses were conducted to evaluate different types of deforestation trends within the country

with respect to roads, county boundaries, and land use designations. If significant trends were identified,

this information could be used to further stratify both activity data and emission factors. Although only

35% of total forest is within 1 km of the nearest road, over half the forest loss occurs in this area. While

7% of the forest areas are more than 5 km away from the road, less than 1% of the forest lost occurs more

than 5 km from a road (Table 6). The dominance of the two forest types at different distances from access

points also does not dramatically change, with 81% of the forest area with >80% tree cover within 1 km of

the road and 91% between 2-5 km from the nearest road. It is not surprising that the deforestation rate is

lower farther away from access points, but in reality very small areas of Liberia are inaccessible and thus

‘accessibility’ does not seem to have a very large influence on deforestation rates.

0

20,000

40,000

60,000

80,000

100,000

120,000

2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014

Hec

tare

s

Forest 30-80%



41

Table 6. Historical mean deforestation rates for various distances from nearest road (historical mean for reference period 2005-

2014)

Distance from road (km)

Pre-deforestation cover

Historical mean deforestation rate

(2005-2014)

% of Total National

Forest Area

<1 Forest >80% 0.46% 28%

Forest 30-80% 0.96% 7%

Total forest 35%

1 - 2 Forest >80% 0.33% 24%

Forest 30-80% 0.89% 4%

Total forest 28%

2 - 5 Forest >80% 0.16% 27%

Forest 30-80% 0.68% 3%

Total forest 30%

>5 Forest >80% 0.02% 7%

Forest 30-80% 0.37% 0.3%

Total forest 7.3%

All Forest Forest >80% 0.37% 86%

Forest 30-80% 1.12% 14%

There are some geographic differences in deforestation rates, with the southern counties having the

highest rates and the northwest and southeast having the lowest rates. These regions also have the most

protected or proposed protected land. Four counties account for a total of 56% of all forest loss: Bong,

Grand Bassa, Lofa, and Nimba, with the remaining counties each accounting for less than 6% of forest loss

(Table 7).

Table 7. Historical deforestation rates (2005-2014) by county and forest cover, grouped by region.

County Pre-deforestation cover


NORTHWEST

Gbarpolu Forest >80% 0.22%

Gbarpolu Forest 30-80% 0.89%

Lofa Forest >80% 0.43%

Lofa Forest 30-80% 1.06%

SOUTHWEST

Bomi Forest >80% 0.78%



42

Bomi Forest 30-80% 1.35%

Grand Cape Mount Forest >80% 0.24%

Grand Cape Mount Forest 30-80% 0.86%

CENTRAL

Bong Forest >80% 0.83%

Bong Forest 30-80% 1.05%

Nimba Forest >80% 0.45%

Nimba Forest 30-80% 1.11%

SOUTH

Grand Bassa Forest >80% 0.71%

Grand Bassa Forest 30-80% 1.09%

Margibi Forest >80% 0.94%

Margibi Forest 30-80% 0.96%

Montserrado Forest >80% 1.00%

Montserrado Forest 30-80% 1.14%

Rivercess Forest >80% 0.39%

Rivercess Forest 30-80% 1.18%

SOUTHEAST

Grand Gedeh Forest >80% 0.12%

Grand Gedeh Forest 30-80% 0.37%

Grand Kru Forest >80% 0.21%

Grand Kru Forest 30-80% 0.92%

Maryland Forest >80% 0.34%

Maryland Forest 30-80% 0.98%

River Gee Forest >80% 0.10%

River Gee Forest 30-80% 0.31%

Sinoe Forest >80% 0.19%

Sinoe Forest 30-80% 0.65%

Based on data evaluated, there are over 700,000 hectares of land within oil palm concessions. In 2000,

using the calculation methods described, over 520,000 ha of this area is classified as forest. However, it is

not known what portion of this area was already oil palm. Between 2005-2014, over 47,000 ha of forest

were lost within the concessions (Table 8). This accounts for 16% of the national forest loss over this time

period. These areas have slightly higher rates of deforestation than forest areas outside of concessions if

the entire reference period is examined. However, since 2012 forest loss rates have risen sharply (Table



43

8, Figure 4) but still do not exceed 3% per year. It is not known if this trend will increase further, however

Oil Palm concession owners may have existing plantation establishment plans that could be consulted.

However, this information could not be accessed for this report. Without this information, it may be

difficult to justify using a deforestation rate from 2011-2014 only for oil palm. In addition, the two forest

classes are found in roughly the same proportions as in the rest of Liberia.

Table 8. Historical mean deforestation rates in Oil Palm concessions in comparison to forest areas outside of Oil Palm

concessions during the Reference Period (2005-2015)

Pre-deforestation cover


(2005-2014)

Total Loss (ha)

2011-2014 mean

deforestation rate

Not oil palm concession areas Forest >80% 0.32% 160,723

Forest 30-80% 1.01% 78,208

Oil palm concessions Forest >80% 0.75% 31,848 1.6%

Forest 30-80% 1.49% 15,207 2.55%

All Forest Forest >80% 0.37% 192,571

Forest 30-80% 1.12% 93,415

Figure 4. Forest loss within Oil Palm concessions over time

-

1,000

2,000

3,000

4,000

5,000

6,000

7,000

8,000

9,000

2000 2002 2004 2006 2008 2010 2012 2014

Are

a o

f F

ore

st L

oss

(h

a)

Forest >80%

Fores 30-80%



44

3.4 Deforestation Activity Data development

To estimate net emissions most accurately, it is necessary to identify the land cover/land use following

deforestation, and develop activity data accordingly. This is because although deforestation will result in

the emissions of greenhouse gases, the vegetation in the land use class following deforestation will

sequester and store greenhouse gases over time. Although accurately knowing what land use the forest

land was converted to historically following forest loss is difficult given the available mapping products,

the Metria/GeoVille map can be used to identify the land cover class in 2015 of all areas not in a forest

class within this map. In addition, concession boundaries for oil palm, rubber, and mining plantations are

available, and it was assumed in this analysis that deforestation in those areas resulted in development of

plantations or mines respectively. Therefore, the following post-deforestation land uses were identified:

Shifting cultivation

Oil palm plantation

Rubber plantation

Non-forest mixed vegetation

Mining

Settlements

Land cover classes from the Metria/GeoVille map were assigned to these post-deforestation land uses for

all deforested lands. Any deforested land not within concession boundaries that had tree cover in 2015

(Metria/GeoVille classes Forest >80% cover, Forest 30-80% cover, and <30% cover) was considered shifting

agriculture. Deforested land outside of concession boundaries that was classified as grassland, shrub, or

bare soil in 2015 was considered non-forest mixed vegetation. Lands classified as settlement in 2015

remained as such. Finally, any deforestation occurring within active concessions was classified as the

respective post-deforestation land use: oil palm, rubber, or mining.

It should be noted that these classifications are based on estimates of land use according to land cover,

and therefore there likely will be some misclassifications. For example, those areas classified as non-forest

mixed vegetation are likely a mix of cropland, the cropping cycle of shifting cultivation, grassland, and

other non-forest mixed vegetation land uses. Additionally, there is currently no way to identify areas

where deforestation results in smallholder plantations, so that land use is not included here. Section 6

provides recommendations on improving future estimates of land use. For this reason, Liberia should

prioritize improving the land use classifications; recommendations are provided in Section 6.

The proportion of each land use in 2015 was identified, based on these classifications, and it was assumed

that these proportions remained constant over time. They were therefore applied to the forest loss for



45

each year for each forest cover class. These numbers provide activity data for deforestation (Table 9), and

reflect an IPCC Approach 3, with land cover classes of forest and non-forest.

Table 9. Historical activity data by forest class, showing estimated post-deforestation land use

Forest >80% cover

Post-deforestation

Shifting Cultivation

Oil palm Plantation

Rubber Plantation



Year Area change (ha)

2005 1,691 519 50 1,036 260 4 3,559

2006 5,099 1,565 150 3,125 784 11 10,735

2007 7,003 2,150 206 4,291 1,077 15 14,742

2008 5,273 1,619 155 3,231 811 12 11,100

2009 13,552 4,160 398 8,304 2,084 30 28,528

2010 4,163 1,278 122 2,551 640 9 8,764

2011 6,379 1,958 187 3,909 981 14 13,428

2012 12,871 3,951 378 7,886 1,979 28 27,093

2013 22,656 6,955 665 13,882 3,484 50 47,692

2014 18,762 5,760 551 11,496 2,885 42 39,496

Forest 30-80% cover

Post-deforestation


Oil palm Plantation

Rubber Plantation



Year Area change (ha)

2005 826 274 36 569 139 1 1,845

2006 2,954 981 129 2,033 495 5 6,597

2007 3,744 1,243 163 2,577 628 7 8,361

2008 3,263 1,083 142 2,246 547 6 7,288

2009 6,792 2,255 296 4,676 1,139 12 15,170

2010 1,625 540 71 1,119 273 3 3,630

2011 2,704 898 118 1,861 454 5 6,039

2012 5,163 1,714 225 3,554 866 9 11,530

2013 9,448 3,136 411 6,503 1,585 17 21,100

2014 7,125 2,365 310 4,904 1,195 13 15,912



46

It is important to note that these broad land use classes do not address drivers of degradation. Notably,

while Table 9 above indicates that shifting cultivation and non-forest mixed vegetation are the main post-

deforestation land cover, it does not include pitsawing and charcoal production, which have been

identified by both the R-PP and the REDD+ Strategy (LTS 2016) as common. However, the MRV roadmap

describes chainsaw logging and charcoal production as drivers of degradation. Indeed, these forest uses

generally do not necessarily in conversion to a non-forest cover. These forest activities are therefore

difficult to quantify without additional data, for instance, on volume of timber and charcoal produced.

4. EMISSION FACTORS Emission factors are measures of the emissions and removals of greenhouse gases per unit of activity data,

usually expressed in units of t CO2e ha-1. Emission factors for land use change are generally developed

using estimates of biomass and carbon stocks of the relevant pools and land cover types and calculating

the difference between pre-deforestation forest carbon stocks and post-deforestation carbon stocks to

determine the change in carbon stocks due to deforestation.

4.1 Forest biomass carbon stocks

Various sources of data may be used to estimate forest biomass and develop emission factors. Potential

sources for generating emission/removal factors include:

Carbon measurement inventories including ground measurement, allometric equations and

remote sensing techniques. These rely on allometric models that relate the biomass of trees with

certain measureable morphological features (e.g. diameter and height) to indirectly quantify

aboveground and belowground tree biomass estimates.

Forest or timber inventories that provide data on the number or trees per hectare or the volume

of timber. These rely on biomass expansion factors to estimate aboveground biomass.

Academic and other research studies that have previously produced biomass and/or carbon

estimates.

Appendix D describes Liberia-relevant allometric equations, biomass expansion factors, and available

biomass data. Currently there are no existing data in Liberia that can be used to develop reliable Tier 3,

country-specific emission factors. To develop these, therefore, it would be necessary for Liberia to develop

a forest sampling scheme and undertake a forest inventory (an example sample design is described in the



47

companion document “Guidance for developing an NFI for Forest Carbon Sampling29). It is recommended

that such an inventory be conducted when time and resources allow, in order to develop an estimate of

carbon stocks and emission factors with an acceptable uncertainty level.

At present, however, we must rely on existing global datasets to develop provisional emission factors.

There are three available pantropical maps of aboveground biomass that can be used to produce carbon

stocks for each of the forest cover classes identified in the Metria/GeoVille map: Saatchi et al. (2011),

Baccini et al. (2012), and Avitabile et al. (2015). (See Appendix D for further description.) These datasets

can be used to develop Tier 2 emissions factors, because while they are global datasets, they are derived

using country-specific biomass estimates, as required for Tier 2 emission factors. The carbon stocks for

aboveground biomass from Avitabile are based on actual, though limited, field data from Liberia, used to

weight and average the Baccini and Saatchi maps. They also match most closely with existing data from

Liberia and neighboring countries, and they provide the most realistic differences between forest classes.

They are therefore likely to be the most accurate of the available global datasets. However, we are using

carbon stocks from Baccini et al (2012) to develop provisional emission factors, because they provide

lower estimates and result in a more conservative reference level. This report will describe historical

emissions and a reference level developed using Bacini carbon stocks. However, when country specific

carbon stock data are developed this should be used in place of the provisional emission factors and the

reference emission level should be recalculated. (See Table 10 for carbon stocks from each.)

Table 10. Above ground carbon stocks in Liberia by forest class, based on global datasets, shown in CO2e ha-1

Forest Class Baccini et al. Saatchi et al. Avitabile et al.

t CO2 ha-1

Forest >80% 364 436 566

Forest 30-80% 317 333 365

Forest <30% 291 311 302

29 Walker SM, Goslee, KM, Eickhoff, G, and Morikawa, Y. 2015. Guidance for developing a National Forest Inventory for Forest Carbon Sampling. Winrock International.



48

Appendix D describes existing data that provides a comparison to the global datasets used here. These

include data from Sierra Leone, Guinea Bissau, and from the Wonegizi REDD+ pilot project in Liberia. These

data are similar to, or larger than the Baccini data in Table 10. This indicates that while country specific

data for Liberia will be different from Baccini, they are unlikely to be substantially lower. Regardless, such

data are required to develop Tier 1 emission factors.

Belowground biomass is estimated based on root to shoot ratios developed by Mokany et al (2006), with

belowground biomass equal to aboveground biomass multiplied by 0.235.

Leaf litter and deadwood are estimated as the fraction of live tree biomass following factors used by

CDM30 (Table 11). All deadwood and leaf litter carbon is assumed to be emitted as CO2e in the year of

forest loss.

Table 11. Default factors for leaf litter and deadwood, taken from CDM A/R Methodological Tool23

Carbon Pools Fraction of live

tree biomass

Litter 0.01

Deadwood 0.01

Total biomass carbon stocks are provided in Table 12.

Table 12. Estimated forest carbon stocks for all pools, based on two global datasets.

AGB Mokany BGB

Litter and deadwood (CDM)

Total carbon stocks

tCO2e/ha, based on Baccini AGB

Forest >80% cover 364 85.5 9.0 458.5

Forest 30-80% cover

317 74.5 7.8 399.3

tCO2e/ha, based on Avitabile AGB

Forest >80% cover 566 133.0 14.0 713.0

30 See: A/R Methodological tool: Estimation of carbon stocks and change in carbon stocks in dead wood and litter in A/R CDM project activities Version 03.0. Data/Parameter tables 5 & 6.



49

Forest 30-80% cover

365 85.8 9.0 459.8

4.2 Post deforestation biomass carbon stocks

As described in section 3.4, post-deforestation land use(s) were identified as shifting cultivation, oil palm

plantations, rubber plantations, non-forest mixed vegetation, mining, and settlements. Carbon stocks

were estimated for each land use/cover, and are shown in Table 13.

Carbon stocks for shifting cultivation are based on a field study for croplands in Ghana’s High Forest Zone24,

which reflects all biomass carbon pools in all cropland (currently cropped or in fallow), rice fields, and agro-

forestry systems. These stocks may be an underestimate of carbon stocks for this land use in Liberia.

Numbers for shifting cultivation exclusively would be expected to be higher than an average across all

types of cropland, and shifting cultivation cycles in Ghana typically involve shorter fallow periods (3-5

years) than what might be expected in Liberia which has a lower population density. However, no

appropriate data were found for Liberia.

Oil palm and rubber plantation carbon stocks are derived from studies conducted on tropical tree crop

systems in Ghana (Kongsager et al 2013)31, as no relevant data were found for Liberia. The values represent

time-averaged carbon stocks for a 30-year rotation, based on the results of the Kongsager study, as cited

in a presentation by the same author.

Lands classified as non-forest mixed vegetation are likely a mix of cropland, the cropping cycle of shifting

cultivation, grassland, shrubland and other non-forest mixed vegetation land cover. These carbon stocks

are based on IPCC default values for cropland. This is likely an overestimate of some land cover, such as

grassland, and an underestimate of other cover, such as shrubland.

It is assumed that there is no biomass carbon on areas converted to mines or infrastructure (roads and

settlements).

31 Konsager et al. The carbon sequestration potential of tree crop plantations. Mitigation Adaptation Strategies for

Global Change (2013) 18:1197–1213. Time-averaged results from

http://orbit.dtu.dk/files/55883745/Carbon_Sequestration.pdf




50

Table 13. Post deforestation biomass carbon stocks by land use

Land Use Carbon Stocks Source

Shifting Cultivation 135.7 t CO2e ha-1 PASCO CORPORATION (2013)

Report on Mapping of Forest

Cover and Carbon Stock in

Ghana. Forest Preservation

Project.32

Oil Palm Plantations 110.0 t CO2e ha-1 Kongsager et al. 2013.

Rubber Plantations 275 t CO2e ha-1 Kongsager et al. 2013.

Non-forest mixed vegetation 18.3 t CO2e ha-1 Based on Ch. 3 LUCF TABLE

3.3.8, value for annual

cropland

Mining and Infrastructure 0 Assumed based on removal of

all vegetation

4.3 Soil carbon stocks

Soil carbon stocks were sourced from the Harmonized World Soil Database 33 . Soil organic carbon

calculations are based on the carbon that is contained in the top 50 cm of the soil. Average values of SOC

were established by comparing forested areas of Liberia to the Harmonized World Soil Database34. The

values are 46.5 t C ha-1 (170.34 t CO2e ha-1) for forest > 80% canopy cover and 44.6 t C ha-1 (163.48 t CO2e

ha-1) for forest with 30-80% canopy cover. The amount of soil carbon emitted as CO2e is a function of land-

use practices that follow forest loss. The IPCC provides guidelines for calculating soil emissions based on

default factors related to the post-deforestation land-use type, management regime, and application of

32 Data were derived from the Forest Preservation Program (FPP), which conducted the Mapping of Forest Cover and Carbon Stock in Ghana project. 33 http://webarchive.iiasa.ac.at/Research/LUC/External-World-soil-database/HTML/ 34 FAO/IIASA/ISRIC/ISSCAS/JRC, 2012. Harmonized World Soil Database (version 1.2). FAO, Rome, Italy and IIASA, Laxenburg, Austria



51

organic materials such as manure35. This IPCC method estimates changes in soil carbon stocks based on

soil factors that account for how the soil is tilled, management practices, and inputs for post deforestation

land use, based on the equation:

ΔSOC = Csoil – (Csoil × FLU × FMG × FI)

Where:

ΔSOC = Soil carbon emitted, t C ha-1

Csoil = Carbon stock in soil organic matter pool (to 30 cm depth), t C ha-1

FLU = Stock change factor for land-use systems for a particular land-use, dimensionless (IPCC

AFOU GL)

FMG = Stock change factor for management regime, dimensionless (IPCC AFOLU GL)

FI = Stock change factor for input of organic matter, dimensionless (IPCC AFOLU GL)

This study assumes that all areas converted to agriculture will be cultivated for at least 20 years with

moderate organic inputs. A summary of SOC stock change factors is given in Table 14. These emissions

from soil respiration are assumed to occur over a 20 year period. However, for the purposes of the

accounting in this study, all soil emissions are considered to occur in the year of forest loss.

Table 14. Change in Soil Organic Carbon calculations, based on IPCC default factors by post-conversion land use

Stratum SOC stock (t CO2e/ha)

IPCC Factors SOC stock at 20 yr

(t CO2e/ha)

Change in Soil C

(t CO2e/ha) FLU FMG FI

Forest > 80% Canopy Cover

170.34


0.80 1.00 1.00 136.27 34.07

Plantations 0.82 1.00 0.92 128.51 41.84

Non-forest mixed vegetation*

0.48 1.00 1.00 81.76 88.58

Mining 0.48 1.00 0.92 75.22 95.12

Infrastructure 0.82 1.00 0.92 128.51 41.84

Forest 30-80% Canopy Cover

163.48

35 Intergovernmental Panel on Climate Change: 2006 IPCC Guidelines for National Greenhouse Gas Inventories. Volume 4: Agriculture, Forestry and Other Land Use - Table 5.5



52


0.80 1.00 1.00 130.79 39.56

Plantations 0.82 1.00 0.92 123.33 47.01

Non-forest mixed vegetation*

0.48 1.00 1.00 78.47 91.87

Mining 0.48 1.00 0.92 72.19 98.15

Infrastructure 0.82 1.00 0.92 123.33 47.01

*Used values for conversion to cropland for non-forest mixed vegetation

4.4 Emission factors

Emission factors for deforestation were calculated separately for each forest class, based on the estimated

land use following deforestation (Table 15), using the following equation:

𝐸𝐹𝑑𝑒𝑓𝑜𝑟𝑒𝑠𝑡𝑎𝑡𝑖𝑜𝑛,𝑥 = 𝐶𝑓𝑜𝑟𝑒𝑠𝑡 − 𝐶𝑑𝑒𝑓𝑜𝑟𝑒𝑠𝑡𝑎𝑡𝑖𝑜𝑛 + 𝐶𝑠𝑜𝑖𝑙

Where:

EFdeforestation,x = Emission factor for deforestation in forest class x; t CO2 ha-1

Cforest,x = Carbon stock in forest class x; t CO2 ha-1

Cdeforestation,y = Carbon stock in post-deforestation land use y; t CO2 ha-1

Csoil,x,y = Change in soil organic carbon, forest class x converted to land use y; t CO2 ha-1

These reflect provisional emission factors for deforestation, and priority should be placed on improving

them using country-specific data.

Table 15. Deforestation emission factors by forest class and post-deforestation land use, using forest carbon stock data from

Baccini et al (2012)

Stratum EF (t CO2e ha-1)

Shifting cultivation

Oil palm plantation

Rubber plantation


Mining Settlement

Forest > 80% Canopy Cover 356.9 390.4 225.4 528.8 553.7 500.4

Forest 30-80% Canopy Cover 303.2 446.3 171.3 472.9 497.5 446.3



53

5. REFERENCE LEVEL DEVELOPMENT

5.1 Historical Emissions

Historical emissions were estimated as the product of activity data and emission factors (Table 16 and

Figure 5).

Table 16. Historical emission estimates for Reference Period, by forest class and post-deforestation land use, based on emission

factors from Avitabile et al dataset.

Forest >80% cover


Oil palm Plantations

Rubber Plantations




2005 603,365 202,589 11,187 547,780 143,925 1,871 1,510,717

2006 1,819,976 611,084 33,746 1,652,311 434,132 5,645 4,556,892

2007 2,499,395 839,209 46,343 2,269,138 596,198 7,752 6,258,035

2008 1,881,996 631,908 34,896 1,708,618 448,926 5,837 4,712,180

2009 4,836,709 1,623,997 89,681 4,391,128 1,153,734 15,001 12,110,249

2010 1,485,816 498,885 27,550 1,348,936 354,422 4,608 3,720,217

2011 2,276,653 764,420 42,213 2,066,916 543,066 7,061 5,700,329

2012 4,593,477 1,542,328 85,171 4,170,304 1,095,714 14,247 11,501,241

2013 8,085,815 2,714,932 149,925 7,340,910 1,928,766 25,078 20,245,427

2014 6,696,246 2,248,364 124,160 6,079,355 1,597,302 20,768 16,766,195

Forest 30-80% cover



Rubber Plantations




2005 250,411 122,380 6,157 268,864 68,930 664 717,406

2006 895,481 437,638 22,017 961,471 246,497 2,375 2,565,480

2007 1,135,055 554,722 27,907 1,218,700 312,444 3,010 3,251,839

2008 989,389 483,533 24,326 1,062,300 272,347 2,624 2,834,519

2009 2,059,278 1,006,407 50,631 2,211,032 566,854 5,461 5,899,663

2010 492,738 240,810 12,115 529,049 135,635 1,307 1,411,654

2011 819,784 400,644 20,156 880,197 225,661 2,174 2,348,615



54

2012 1,565,213 764,948 38,484 1,680,557 430,853 4,151 4,484,206

2013 2,864,344 1,399,858 70,425 3,075,425 788,463 7,596 8,206,110

2014 2,160,081 1,055,671 53,109 2,319,263 594,601 5,728 6,188,454

Sum of All Forests



Rubber Plantations




2005 853,775 324,969 17,344 816,644 212,855 2,535 2,228,122

2006 2,715,457 1,048,722 55,763 2,613,782 680,629 8,019 7,122,372

2007 3,634,450 1,393,931 74,250 3,487,838 908,643 10,762 9,509,874

2008 2,871,385 1,115,441 59,221 2,770,917 721,273 8,461 7,546,699

2009 6,895,987 2,630,404 140,312 6,602,159 1,720,588 20,462 18,009,913

2010 1,978,554 739,695 39,664 1,877,985 490,057 5,915 5,131,871

2011 3,096,437 1,165,064 62,369 2,947,113 768,726 9,235 8,048,944

2012 6,158,690 2,307,277 123,655 5,850,861 1,526,567 18,397 15,985,448

2013 10,950,159 4,114,790 220,350 10,416,335 2,717,229 32,674 28,451,538

2014 8,856,326 3,304,035 177,270 8,398,617 2,191,904 26,497 22,954,649

Figure 5. Historical emission estimates based on emission factors from Baccini et al dataset.

-

5

10

15

20

25

30

1 2 3 4 5 6 7 8 9 10

tCO

2e

Mil

lio

ns

Forest 30-80% Forest >80% All Forest



55

5.2 Projecting Future Emissions without REDD

While Liberia’s historical emissions show a general upward trend over time, the FCPF requires use of the

historical average. According to the Carbon Fund Methodological Framework, Indicator 13.2:

The Reference Level does not exceed the average annual historical emissions over the Reference Period, unless the ER Program meets the eligibility requirements in Indicator 13.2. If the available data from the National Forest Monitoring System used in the construction of the Reference Level shows a clear downward trend, this should be taken into account in the construction of the Reference Level.

Liberia’s average historical emissions over the reference period 2005-2014 are 12,498,943t CO2e/yr based

on Baccini et al data and the average annual historical deforestation rates for each forest class. The

cumulative greenhouse gas emissions over ten years would then be over 125 million t CO2e (Figure 6). This

represents the Reference Emission Level, without adjustments for national circumstances.

Figure 6. Total historical emission estimates over 2005-2014 reference period and historical average projected into future.

0

5

10

15

20

25

30

20

05

20

06

20

07

20

08

20

09

20

10

20

11

20

12

20

13

20

14

20

15

20

16

20

17

20

18

20

19

20

20

Em

issi

on

s t

CO

2e

yr-

1

Mil

lio

ns



56

5.3 Adjusting for national circumstances

According to the Carbon Fund Methodological Framework, Indicator 13.2:

The Reference Level may be adjusted upward above average annual historical emissions if the ER

Program can demonstrate to the satisfaction of the Carbon Fund that the following eligibility

requirements are met:

i.) Long-term historical deforestation has been minimal across the entirety of the country, and

the country has high forest cover;

ii.) National circumstances have changed such that rates of deforestation and forest degradation

during the historical Reference Period likely underestimate future rates of deforestation and

forest degradation during the Term of the ERPA.

If a country meets these criteria, the Carbon Fund allows the Reference Level to be based on an adjustment

of the average annual historical emissions over the Reference Period, not to exceed 0.1% of carbon stocks

(Indicator 13.4). Using the carbon stocks from Baccini, 0.1% is equal to 2,844,633 t CO2e, which serves as

a cap on upward adjustment of the reference level.

Liberia arguably meets these criteria, with relatively low historical levels of deforestation36, remaining

forests over a large percent of the country, and substantially changed national circumstances. Since the

end of the civil war, the economy has grown steadily, especially during the Historical Reference Period

(2005-2014). The Ebola outbreak paused this growth, yet now that this crisis has ended, the economy is

expected to rebound quickly (Figure 7).

36 It is important to note, however, that Liberia does not fit the definition of a “High Forest, Low Deforestation” country, which must have >50% forest cover and a deforestation rate of <0.22% per year. While Liberia has more than 50% forest cover based on Metria Geoville’s mapping analysis, the deforestation rate is higher than 0.22% annually.



57

Figure 7. Liberia's GDP over time (World Bank37)

As described in the economic analysis of the drivers of forest change in Liberia, provided in Annex 1, prior

to the civil wars in Liberia there was substantial production in the country, particularly in timber, iron ore,

and diamonds. In all cases, production dropped off significantly during and after the wars, and has begun

to increase again in recent years in the case of timber and iron ore. This likely indicates additional land use

change as a result of conversion of areas for mining, removal of forest cover for timber production, and

impacts of associated activities such as road building.

Based on existing information, it is extremely likely that the economy of Liberia will continue to recover at

its pre-Ebola growth rate of around 7% (Figure 8).

37 http://data.worldbank.org/indicator/NY.GDP.MKTP.KD.ZG/countries

0

0.5

1

1.5

2

2.5

1985 1990 1995 2000 2005 2010 2015

GD

P (

$)

Bil

lio

ns



58

Figure 8. GDP Annual Growth Rate (%) (World Bank)

Additionally, the economic analysis shows a steady increase in production of palm oil, starting in 1966 and

continuing into the future. This sector of the economy is likely to have an even larger impact than indicated

by this analysis, as evidenced by the extent of land currently allocated for oil palm concessions, but as yet

undeveloped. According to the land use analysis conducted under Liberia’s REDD Readiness efforts (LTS

International 2016) concessions have been granted to four international palm oil companies since 2009,

covering over 620,000 hectares. As discussed in Section 3.3, across the Reference Period, over 47,000

hectares of forest have been lost within existing palm oil concession boundaries, accounting for

approximately 7% of concession areas. Over the next 15 years, however, it is anticipated that between

160,000 and 352,000 hectares of forest area will be cleared for oil palm plantations. Additional efforts

could be made to work with the palm oil companies to determine their expected conversion plans in the

future. If sufficient evidence is made available, it may be possible for this to be used to justify adjusting

the Reference Emission Level.

While existing evidence indicates that there is justification for an upward adjustment of the historical

average emissions, there must also be justification for the numerical adjustment applied, and such

justification requires additional data and analysis. While the economic analysis indicates increasing

development relative to recent past trends, there was not a correlation found between mining, forestry,

and palm prices and deforestation rates in the country, making it difficult to quantify the impact of

anticipated future price increases and related expanded development. One possible reason for the

difficulty in making the link between prices and deforestation has to do with the lack of information on

the specific activities that lead to instances of deforestation, e.g., mining, oil palm development,

0

1

2

3

4

5

6

7

8

9

10

2005 2007 2009 2011 2013 2015

GD

P G

row

th (

%)



59

agriculture. If improved data on the drivers of deforestation were available over time, we could attempt

to develop land use change models. Population was correlated with deforestation rates, and it is likely

that there is a correlation between GDP growth and deforestation. Additional analysis could be conducted

to quantify expected increase in deforestation based on this correlation.

The LTS International draft Land Use Analysis (2016) does provide some indication of anticipated land use

change in the near future. This report assessed spatial data on land use from the Government of Liberia,

which is focused on concessions for forestry, agriculture, and mining. Based on their assessment, a

significant proportion of land is under threat for development in the near future. Table 17 provides the

estimates of forest area the land use analysis projects will be impacted, by activity.

Table 17. Area of development by activity, based on land use analysis by LTS International (2016); only those activities likely

to result in deforestation are included

Activity Area (ha)

>80% cover

30-80% cover

total

Projected oil palm expansion* 180,810 180,810

Timber sale expansion 94,981 52,432 147,413

Mining – Mineral Development Agreements 134,042 66,508 200,550

Mining – class A 81,596 55,649 137,245

TOTAL DEFORESTATION 310,619 355,399 666,018 *Oil palm expansion was not divided by land cover, so it was conservatively assumed all the land falls under the 30-80% cover

Projected activities that would result in deforestation include oil palm expansion, forest management

contracts, and mining. Combined, these three activities account for 1.167 million hectares, as projected in

the land use analysis. Using the emission factors based on Baccini carbon stocks, and averaging over 20

years, this represents annual emissions of 16,794,096 t CO2e. Therefore, such forest loss would result in

greater emissions than the historical average emissions (12,498,943 t CO2e) plus the 0.1% cap (2,844,633

t CO2e) specified by FCPF for an upward adjustment of the reference level, totalling emissions of

15,343,576 t CO2e.

While these estimates provide a general sense of potential future emissions, and may be justification for

Liberia arguing that it should be allowed to adjust its REL to the extent of the FCPF cap, it is recommended

that additional documentation of expansion plans be developed through consultation with the existing

concession companies and other stakeholders. Based on the information compiled, discussions should be

held with FCPF on options for adjusting the Reference Emission Level based on National Circumstances.



60

5.4 Reference Emission Level

The Reference Emission Level described here for Liberia is based on Reference Period of 2005-2014. It

currently accounts for only emissions from deforestation, based on data available for the country. Activity

data were developed using deforestation data developed by Hansen et al (2013) and adjusted for Liberia

based on the 2015 landcover map produced by Metria/Geoville. Post deforestation land uses were

extrapolated based on land cover in the Metria/Geoville map; improved land use classification is strongly

recommended to improve the REL. Emission factors were developed using Baccini et al (2012) global

biomass data, and should be considered provisional; it is recommended that Liberia collect country-

specific biomass data in order to develop Tier 2/3 emission factors.

Liberia must decide whether to propose a reference level based strictly on average historical emissions or

based on an adjustment for national circumstances. However, FCPF must permit Liberia to adjust for

national circumstances, and it is not clear that such an adjustment would be allow. An initial suggested

adjustment is based on justification using the draft LTS Land Use Analysis report (2016). The two options

for Liberia’s Reference Emission Level, given the assumptions described here, are provided in Table 18.

Table 18. Potential Reference Emission Levels with average historical emissions and adjusted for national circumstances

Reference Emission Level tCO2e

Based on average historical emissions 12,498,943

Adjusted for national circumstances 15,343,576

6. NEXT STEPS AND RECOMMENDATIONS

6.1 Submitting the proposed REL

Sections 1-5 of this report are intended to provide the background, methods, rationale, and findings for a

proposed Reference Emission Level that can be submitted to the UNFCCC and/or the World Bank Carbon

Fund. It is also intended that this REL can meet requirements established under a multilateral agreement

between Norway and Liberia, although such requirements have not yet been fully specified. The REDD+

Implementation Unit of Liberia’s Forestry Development Authority, along with the REDD+ Technical

Working Group must decide the appropriate venue for submission of this proposed REL.



61

Submitting a RL/REL to such bodies inevitably entails some level of back and forth regarding methods used,

decisions that have been made, and the appropriateness of the recommended REL.

Guidelines and procedures regarding the UNFCCC’s technical assessment of submitted proposed RL/RELs,

with regard to use for results based payments, can be found in Decision 13/CP.1938. The objectives of the

technical assessment, as described in the annex of Decision 13/CP.19, are:

“(a) To assess the degree to which information provided by Parties is in accordance with the

guidelines for submissions of information on forest reference emission levels and/or forest reference

levels contained in the annex to decision 12/CP.17 for the construction of the forest reference

emission levels and/or forest reference levels;

(b) To offer a facilitative, non-intrusive, technical exchange of information on the construction of

forest reference emission levels and/or forest reference levels with a view to supporting the capacity

of developing country Parties for the construction and future improvements, as appropriate, of their

forest reference emission levels and/or forest reference levels, subject to national capabilities and

policy.”

6.2 Future improvements

Given that a step-wise approach has been recommended, there are a number of improvements that

Liberia can make to the proposed REL, as resources allow. These have been mentioned throughout this

report, and are summarized here. Additional capacity building activities are described in further detail in

Annex 2.

A. Develop a Forest Carbon Monitoring System

Because Liberia does not have a National Forest Inventory or similar system, global datasets have been

used here to estimate forest biomass and develop emission factors. These are, at best, IPCC Tier 2 emission

factors, and should be improved by developing country-specific data on Liberia’s forest and forest carbon

stocks. This can be done by implementing a National Forest Inventory or a smaller scale Forest Carbon

Monitoring System. Guidance on developing such an inventory, conducting measurements, and analysing

the resulting data are provided in the following documents, presented as companions to this report:

Walker SM, Goslee, KM, Murray, L, Eickhoff, G, and Morikawa, Y. 2016. Guidance on Developing a

National Forest Inventory for Forest Carbon Sampling. Adapted by Winrock International.

38 http://unfccc.int/resource/docs/2013/cop19/eng/10a01.pdf#page=34



62

Walker, SM, TRH Pearson, FM Casarim, N Harris, S Petrova, A Grais, E Swails, M Netzer, KM Goslee

and S Brown. 2016. Standard Operating Procedures for Terrestrial Carbon Measurement: Version

2016. Winrock International.

Goslee, K, SM Walker, A Grais, L Murray, F Casarim, and S Brown. 2015. LEAF Technical Guidance Series for the Development of a Forest Carbon Monitoring System for REDD+, Module C-CS: Calculations for Estimating Carbon Stocks. Winrock International.

B. Develop accurate data on land use

The RL and any developed MRV system would benefit greatly from better data on Liberia’s land area under

different land use types, for example, plantations, active logging concessions, active fuelwood

extraction/charcoal, swidden agriculture, cocoa, and others. Some such data exists in Liberia; for example

there are shapefiles of logging concession boundaries and plantations in Liberia, but it is unclear how

accurate these data are and the areas in active concession are much smaller than the total concession

boundaries. With land uses such as smallholder agriculture and plantations, fuelwood, and charcoal

extraction, the government is not involved and typically activities are performed by individual farmers and

families, not by companies, making any reporting of areas very difficult. However, accurate identification

of land uses and land use change improves total emission estimates. It is important for Liberia to maintain

data on active plantation areas, including both commercial and smallholder. It is also important to identify

areas of permanent agriculture and areas of shifting cultivation, regardless of whether the current

landcover is cropped or fallow. Liberia can undertake improved land use classification by identifying post-

deforestation land uses, rather than just land cover. Land use classification can also be done using remote

sensing methods, described in brief in Appendix F.

C. Develop data required for justification for adjusting REL

The options for adjusting the REL for national circumstances described in section 5.3 are likely to require

additional justification to be accepted by FCPF. This will require additional data and information. Specific

planned or expected development activities should be identified and documented, especially from oil

palm plantations. This information can be used to estimate emissions that would result from such activities

and develop a defensible quantitative adjustment of average historical emissions, for a reference emission

level based on national circumstances.

D. Improving Activity Data estimates for deforestation

While the methods that have been used to develop estimates of deforestation activity data described in

this report are appropriate for developing a Reference Emission Level, they could be improved if time and

resources allow. An initial improvement would entail conducting a complete ground truthing exercise for



63

the Metria/Geoville 2014/15 Landcover Map. This would require sampling across the country, in areas

representing all major land cover types and land cover changes. The methods applied during the limited

ground truthing exercise (described in Appendix B) would be appropriate for such an exercise. If biomass

data are collected, either in a forest carbon monitoring system, or in conjunction with ground-truthing

efforts, L-band radar data would be used to validate deforestation estimates going forward. Biomass plot

data could enable calibration of maps, by assisting in determination of whether a change in radar

backscatter is sufficient to be deforestation or degradation, and establishments of thresholds for

forest/non-forest.

Additionally, improved methods could be implemented that would increase the accuracy of activity data

estimates over the historical reference period (Appendix E). This could potentially increase the accuracy

of the deforestation area estimates for each forest stratum. However, it is not possible to estimate the

expected accuracy increase. The costs for undertaking such improvements for historical activity data are

significantly high and thus likely cost prohibitive. However, in the future under an MRV system, improved

methods to monitor deforestation events overtime are recommended. Potential improvements for both

historical AD and the future MRV, including cost estimates, are described in Appendix E.

E. Estimate emissions from Degradation and Enhancement

In order to move from the current Reference Emission Level to a full Reference Level, accounting for all

significant sources of emissions from land use, as well as removals, Liberia will need to include degradation

and enhancement in its accounting. This will require increased data collection capacity and activities.

Options for including degradation are described in Annex 1. This should be pursued in the medium term,

following implementation of a National Forest Inventory. Incorporating removals from enhancement into

the Reference Level and a future MRV should be a longer term objective, if Liberia decides it is appropriate.

It would be reasonable and beneficial to include enhancement if the country intends to undertake

activities to substantially increase the extent of forest land and/or canopy cover of existing forests. At

present, there is no indication that such activities will be undertaken in the near future, nor does Liberia

have the capacity to gather the necessary data to allow appropriate accounting of enhancement.

F. Progressing from REL to MRV

As described in Box 1, the Reference Emission Level (REL) provides estimates of historical emissions and a

projection of those emissions into the future, in the absence of a REDD+ program, while a system for

Monitoring, Reporting, and Verification (MRV) is needed to compare actual emissions with the REL, and

estimate emission reductions. The MRV should follow the same methods as the REL, although it is usually

possible to improve the methods used for data collection under the MRV. The MRV roadmap provides an



64

initial description of what steps Liberia can undertake to implement a full MRV system. As described in

Annex 2, many of these steps are currently underway.



65

7. REFERENCES Asante, W. and N. Jengre. 2010. Wonegizi REDD+ Project, Lofa County, Liberia: Report on Initial Carbon

Stock Assessment and Capacity Building in Carbon Accounting. Report produced by NCRC, Katoomba Group, and Forest Trends.

Amaro et al. 2012. World Bank Land Use Change Analysis for REDD Baselines Scenario Definition & Carbon Stock Assessment for REDD PROJECT in Guinea Bissau - UPDATE AND REVISION Appendix III.

Avitabile, V., M. Herold, G. B. M. Heuvelink, S. L. Lewis, O. L. Phillips, G. P. Asner, J. Armston, P. Asthon, L. F. Banin, N. Bayol, N. Berry, P. Boeckx, B. de Jong, B. DeVries, C. Girardin, E. Kearsley, J. A. Lindsell, G. Lopez-Gonzalez, R. Lucas, Y. Malhi, A. Morel, E. Mitchard, L. Nagy, L. Qie, M. Quinones, C. M. Ryan, F. Slik, T. Sunderland, G. Vaglio Laurin, R. Valentini, H. Verbeeck, A. Wijaya, and S. Willcock. 2015. An integrated pan-tropical biomass map using multiple reference datasets. Global Change Biology:n/a-n/a.

Baccini, A., S. J. Goetz, W. S. Walker, N. T. Laporte, M. Sun, D. Sulla-Menashe, J. Hackler, P. S. A. Beck, R. Dubayah, M. A. Friedl, S. Samanta, and R. A. Houghton. 2012. Estimated carbon dioxide emissions from tropical deforestation improved by carbon-density maps. Nature Clim. Change 2:182-185.

Bayol, N. and J.F. Chevalier. 2004. Current state of the forest cover in Liberia – Forest information critical to decision making. Forest Resource Management Final Report to the World Bank. Forest Resource Management, Washington.

Christie, T., M.K. Steininger, D. Juhn. and A. Peal. 2007. Fragmentation and clearance of Liberia's forest during 1986-2000. Oryx Vol 41 No. 4.

Collins, M.B., & Mitchard, E.T.A. 2015. Integrated radar and lidar analysis reveals extensive loss of remaining intact forest on Sumatra 2007-2010.Biogeosciences, 12, 6637-6653, DOI: 10.5194/bg-12-6637-2015

Ebeling, J. and R.A. Asare. 2011. Wonegizi REDD Project Feasibility Assessment. Report prepared by Forest Trends and the West Africa Katoomba Incubator

Fauna & Flora International and RSS GmbH. 2014. Wonegizi Community-based REDD+ and Protected Area Project, Technical Specification.

Forest Carbon Partnership Facility (FCPF). 2013. FCPF Carbon Fund Methodological Framework. Available at https://www.forestcarbonpartnership.org/carbon-fund-methodological-framework.

Goslee, K.M., S. Brown, S.M. Walker, L. Murray, and T. Tepe. 2015. Review of aboveground biomass estimation techniques. High Carbon Stock Science Study, Consulting Study 3.

Goslee, K., E. Mitchard, A. Grais, M. Netzer, S. Walker, J. Donovan, P. Mulbah. 2016. Assistance for Development of Reference Scenario for REDD+ Readiness: Interim Report for Republic of Liberia Forest Development Authority.

http://dx.doi.org/10.5194/bg-12-6637-2015

http://dx.doi.org/10.5194/bg-12-6637-2015

https://www.forestcarbonpartnership.org/carbon-fund-methodological-framework



66

Hansen M.C., Potapov P.V., Moore R., Hancher M., Turubanova S.A., Tyukavina A., Thau D., Stehman S.V., Goetz S.J., Loveland T. R., Kommareddy A., Egorov A., Chini L., Justice C.O., Townshend J.R.G. (2013) High-Resolution Global Maps of 21st-Century ForestCover Change. Science 342(6160) pp. 850-853 DOI: 10.1126/science.1244693

Hess, P. and S. Trainer. 2006. Forest Inventory in Liberia, Results and Interpretation. Report for the World Bank Group by Deutsche Forstservice GmbH.

Joshi, N.P., Mitchard, E.T.A., Woo, M., Torres, J., Moll-Rocek, J., Ehammer, A., Collins, M., Jepsen, M.R., & Fensholt, R. 2015. Mapping dynamics of deforestation and forest degradation in tropical forests using radar satellite data. Environmental Research Letters 10, 3 DOI: 10.1088/1748-9326/10/3/034014

JV Metria & GeoVille. 2016. Liberia Forest Classification, V 1. Mapping performed under Forestry Development Authority (FDA) contract No: FDA/FCPF/JVMG/LLCFM/01/14.

Konsager et al. The carbon sequestration potential of tree crop plantations. Mitigation Adaptation Strategies for Global Change (2013) 18:1197–1213. Time-averaged results from: http://orbit.dtu.dk/files/55883745/Carbon_Sequestration.pdf

LTS International. 2016. Land Cover and Land Use Analysis, Draft Report (DR-2a). Submitted to the Liberian Forestry Development Authority by LTS International Limited and NIRAS, 31 March 2016.

Mitchard, E.T.A., S.S. Saatchi, I.H. Woodhouse, G. Nangendo, N.S. Ribeiro, M. Williams, C.M. Ryan, S.L. Lewis, T.R. Feldpausch, & P. Meir. 2009. Using satellite radar backscatter to predict above-ground woody biomass: A consistent relationship across four different African landscapes. Geophysical Research Letters, 36, L23401, doi:10.1029/2009GL040692

Mitchard, E., S. Saatchi, A. Baccini, G. Asner, S. Goetz, N. Harris, and S. Brown. 2013. Uncertainty in the spatial distribution of tropical forest biomass: a comparison of pan-tropical maps. Carbon Balance and Management 8:10.

Mitchard, E. T. A., T. R. Feldpausch, R. J. W. Brienen, G. Lopez-Gonzalez, A. Monteagudo, T. R. Baker, S. L. Lewis, J. Lloyd, C. A. Quesada, M. Gloor, H. ter Steege, P. Meir, E. Alvarez, A. Araujo-Murakami, L. E. O. C. Aragão, L. Arroyo, G. Aymard, O. Banki, D. Bonal, S. Brown, F. I. Brown, C. E. Cerón, V. Chama Moscoso, J. Chave, J. A. Comiskey, F. Cornejo, M. Corrales Medina, L. Da Costa, F. R. C. Costa, A. Di Fiore, T. F. Domingues, T. L. Erwin, T. Frederickson, N. Higuchi, E. N. Honorio Coronado, T. J. Killeen, W. F. Laurance, C. Levis, W. E. Magnusson, B. S. Marimon, B. H. Marimon Junior, I. Mendoza Polo, P. Mishra, M. T. Nascimento, D. Neill, M. P. Núñez Vargas, W. A. Palacios, A. Parada, G. Pardo Molina, M. Peña-Claros, N. Pitman, C. A. Peres, L. Poorter, A. Prieto, H. Ramirez-Angulo, Z. Restrepo Correa, A. Roopsind, K. H. Roucoux, A. Rudas, R. P. Salomão, J. Schietti, M. Silveira, P. F. de Souza, M. K. Steininger, J. Stropp, J. Terborgh, R. Thomas, M. Toledo, A. Torres-Lezama, T. R. van Andel, G. M. F. van der Heijden, I. C. G. Vieira, S. Vieira, E. Vilanova-Torre, V. A. Vos, O. Wang, C. E. Zartman, Y. Malhi, and O. L. Phillips. 2014. Markedly divergent estimates of Amazon forest carbon density from ground plots and satellites. Global Ecology and Biogeography 23:935-946.

http://dx.doi.org/10.1088/1748-9326/10/3/034014

http://dx.doi.org/10.1088/1748-9326/10/3/034014


http://onlinelibrary.wiley.com/doi/10.1029/2009GL040692/full



67

Mokany, K., R.J. Raison, and A.S. Prokushkin. 2006. Critical analysis of root:shoot ratios in terrestrial biomes. Global Change Biology 12: 84-96.

Netzer M. and S. Walker. 2013. Gola REDD Project Baseline Report Application of VM0007 BL-UP. Developed for the Royal Society for the Protection of Birds by Winrock International.

Radelet, S. 2007. Reviving Economic Growth in Liberia. Center for Global Development, Working Paper Number 133.

Ryan, C.M., Hill, T., Woollen, E., Ghee, C., Mitchard, E.T.A., Cassells, G., Grace, J., Woodhouse, I.H., & Williams, M. 2012. Quantifying small-scale deforestation and forest degradation in African woodlands using radar imagery. Global Change Biology, 18, 243-257. DOI: 10.1111/j.1365-2486.2011.02551.x

Saatchi, S. S., N. L. Harris, S. Brown, M. Lefsky, E. T. A. Mitchard, W. Salas, B. R. Zutta, W. Buermann, S. L. Lewis, S. Hagen, S. Petrova, L. White, M. Silman, and A. Morel. 2011. Benchmark map of forest carbon stocks in tropical regions across three continents. Proceedings of the National Academy of Sciences 108:9899-9904.

Walker et al. 2013, Technical Guidance on Development of a REDD+ Reference Level. Winrock International Developped for the USAID Lowering Emissions in Asia’s Forest Program.

http://onlinelibrary.wiley.com/doi/10.1111/j.1365-2486.2011.02551.x/abstract

http://onlinelibrary.wiley.com/doi/10.1111/j.1365-2486.2011.02551.x/abstract



68

APPENDIX A: AVAILABLE LAND COVER, LAND COVER CHANGE,

AND RAW SATELLITE DATA

Developed by Ed Mitchard

This table provides all data identified as potentially relevant to assessing land cover change in Liberia, as

well as a description of how they could be used and whether they were used for analysis.

Name &

description

Year(s) Resolution Utility Source Use in analysis

Land Cover

Globcover

2.2 and 2.3.

Global land

cover maps.

2009 &

2005-6

300 m Differentiates between

some land use and land

cover classes that could be

useful in baseline

production, including

agriculture-forest mosaic.

Resolution probably too

coarse though. Not suitable

for land cover change.

ESA

http://due.esrin.

esa.int/page

_globcover.php

Not used, as 300

m pixel size too

coarse for

requirements.

CCI Land

Cover

Products.

Global land

cover maps.

98-02;

03-07;

08-12

300 m Differentiates between

some land use and land

cover classes that could be

useful in baseline

production, including

agriculture-forest mosaic.

Resolution probably too

coarse though. Reasonably

suitable for land cover

change. In theory better

than Globcover, but less

well known.

ESA

http://www.esa-

landcover-

cci.org/?q=node/

156

Not used, as 300

m pixel size too

coarse for

requirements.

http://due.esrin.esa.int/page_globcover.php



http://www.esa-landcover-cci.org/?q=node/156






69

Name &

description


Liberia land

cover and

land use

map

? ? - vector Could assist with baseline

map

LISGIS via Tera

Tech ARD

Not used; full

dataset never

received and

definitions of

forest and dates

never clear

GeoVille/Me

tria Land

Cover map

2014 ? 5m/30m Essential for baseline, and

first time point for

monitoring.

GeoVille/Metria Extensively used.

Final product was

10 m resolution

and dated 2015.

FRM Forest

cover map

2004 30 m Potentially useful for

baseline. Forest cover map

based on Landsat. Not clear

what forest definition used,

nor what classes

Bayol & Chevalier

2004

Not used due to

lack of clarity on

classes, and actual

spatial data never

made available.

Canopy Cover

Hansen et

al. (2013)

“Tree

Cover”

product.

Percentage

tree cover

for every

pixel.

Global.

2000 30 m Could be used to make a

forest/non-forest map for

2000 – useful for baseline.

Unlike straight landcover

maps, can make a forest

cover map based on

different forest definitions.

Original paper in

Science:

http://www.scie

ncemag.org

/content/342/61

60/850

Raw data can be

downloaded

from

http://earthengi

nepartners.appsp

ot

.com/science-

Used; converted

to <30, >30 and

>80 % canopy

cover classes

using the

GeoVille/Metria

map and the

Hansen

deforestation

product to

ascertain which

areas should not

have changed.

http://www.sciencemag.org/content/342/6160/850




http://earthenginepartners.appspot.com/science-2013-global-forest






70

Name &

description


2013-global-

forest

Forest Loss

Hansen et

al. (2013)

“Forest

Loss”

product.

Global.

Annual

data

from

2000-

2013

30 m Could provide baseline

deforestation data.

Original paper in

Science:

http://www.scie

ncemag.org

/content/342/61

60/850

Raw data can be

downloaded

from

http://earthengi

nepartners.appsp

ot

.com/science-

2013-global-

forest

Used extensively

Forest Gain

Hansen et

al. (2013)

“Forest

Gain”

product.

Global.

Gives

gain

inform

ation

betwee

n 2001-

2013.

30 m Could help with baseline

production, but NB no

annual information, states

gain occurred sometime

between 2001 and 2013.

Original paper in

Science:

http://www.scie

ncemag.org

/content/342/61

60/850

Raw data can be

downloaded

from

http://earthengi

Not useful due to

lack of date

information for

gain




















71

Name &

description


nepartners.appsp

ot

.com/science-

2013-global-

forest

Satellite data

Landsat.

Optical

satellite

data.

1972-

presen

t

30 m (from

mid-80’s)

Data available throughout

2000-2015 from Landsats

5, 7 & 8, though with some

2-3 year gaps due to cloud

cover. Dry season data can

be used to make

reasonably reliable land

cover maps.

Data free to

download from

http://earthexplo

rer.usgs.gov/

Used by G/M

extensively to

create their 2014

map, in

combination with

small areas of 5 m

resolution

RapidEye data.

ALOS

PALSAR. L-

band radar

satellite

data.

2007-

2010

20 m Free mosaics available at

25 m resolution annually

from 2007-2010. These

data have been shown

before to be very useful at

differentiating different

stages of forest

regeneration, particularly

in farmland-regrowth

mosaics. Was used

successfully with Landsat in

neighbouring area of Sierra

Leone. Additional

advantage of seeing

through clouds

http://www.eorc

.jaxa.jp/ALOS/en

/palsar_fnf/fnf_i

ndex.htm

Downloaded and

processed, but

without ground

data creating

maps of

deforestation/deg

radation and

regrowth was not

possible.






http://earthexplorer.usgs.gov/

http://earthexplorer.usgs.gov/

http://www.eorc.jaxa.jp/ALOS/en/palsar_fnf/fnf_index.htm






72

Name &

description


JERS-1. L-

band radar

satellite

data

1996 100 m Free mosaic available for

1996. Could be used to

help with early baseline.

Data similar to PALSAR, but

lower quality.

JAXA. Not

available online,

but EM has data.

Not used as 1996

was seen as too

early to be helpful

for baseline

development, plus

same lack of

ground data

problems as for

ALOS.

Sentinel 1.

C-band

radar

satellite

data.

2014- ~10 m Data free from 2014 and

will continue into 2030’s.

Cloud-free, but shorter

wavelength than L-band so

less useful for

differentiating forest types.

Could be useful for

degradation mapping,

though tricky to analyse.

ESA.

https://sentinel.e

sa.int/web/senti

nel/missions/sen

tinel-1

Not used. Only

available from

2015 in the end,

so too late for

baseline

development and

good map already

existed for 2015

from G/M.

Sentinel 2.

Optical

satellite

data.

2015- 10 m Data will be available every

10 days from late in 2015,

increasing to every 5 days

later this decade. Could be

used for monitoring in the

future.

ESA. Portal not

set up as yet, but

in theory data

will be free and

open

Not available until

early 2016

https://sentinel.esa.int/web/sentinel/missions/sentinel-1






73

APPENDIX B: FIELD ASSESSMENT OF SPATIAL DATASET

ACCURACY As historical ground data in Liberia are especially sparse, an attempt was made to assess the accuracy of

the spatial datasets applied in this study and identify potential shortcomings in accuracy. A field

assessment was undertaken in early 2016 whereby crews visited a number of pre-determined sites and

were tasked with collecting a set of measurements and photos to help corroborate land cover

classifications as determined by the datasets actual observed land cover.

This effort was not designed to be a formal accuracy assessment or ground truthing exercise of the

Geoville/Metria land cover map as it did not include a comprehensive selection of evenly distributed sites.

Rather, instead it was designed to collect independent field data to evaluate the accuracy of the forest

stratification in the G/M land cover map. Due to resource limitations, measurement sites were selected

with a bias toward accessibility.

The three datasets whose accuracy was being assessed have been extensively used in the creation of land

cover and land cover change maps for Liberia in the development of its Reference Levels include:

- The Geoville/Metria landcover map for 2014/15

- The Hansen et al. (2013) Forest Loss product (2000-2014)

- The Hansen et al. (2013) Percent Canopy Cover Map for the year 2000

These datasets invariably contain errors, such as mis-classifications in the G/M landcover map, false

detections of deforestation or missed deforestation events in the Forest Loss product, and incorrect

estimates of canopy cover in the Hansen et al. (2013) Percent Forest Cover map for the year 2000.

Nevertheless, it is generally accepted that global or regional datasets, even with errors, are suitable for

use in producing Reference Levels (or Reference Emission Levels)39,40. However, existing guidance clearly

indicates that they should only be used if verified against ground data, to correct for biases and allow and

estimation of confidence intervals.

39 GFOI Methods & Guidance Document, http://www.gfoi.org/methods-guidance/ 40 GOFC-GOLD Sourcebook of methods and procedures for monitoring, measuring & reporting http://www.gofcgold.wur.nl/redd/index.php


http://www.gofcgold.wur.nl/redd/index.php



74

Methods

The field team visited 70 plots located across three distinct areas of Liberia (see Figure B1). Plots were

random locations chosen in advance, stratified by original G/M landcover type. The field team were given

discretion as to which sites could be skipped based on accessibility. The field team used GPS devices to

locate each of the predetermined sites, and once located, established a 50m circular plot. Information on

the following characteristics were then recorded to describe the area within the circular plot:

- Land cover type

- Crop type if grown

- Canopy cover (estimated using a spherical densitometer)

- Maximum tree height (estimated using a vertex hypsometer)

- 4 photographs, one in each compass direction.



75

Figure B1. Location of ground truth plots

Results

An initial examination of the landcover types suggested the team had over-sampled agriculture and under-

sampled forest types, which was not expected given the equal weighting of the input points (Table B1).

Table B1. Ground truth points by field-reported landcover/landuse type

Landcover class Number of points

Forest (>80% canopy cover) 2

Forest (30-80% canopy cover) 18

Non-forest with trees (1-30%) 4



76

Farmed 42

Swamp 4

Total 70

Land cover versus land use

Liberia’s landscapes are largely comprised of a mosaic of agricultural land at various stages of production

and fallow. Even under somewhat shortened fallow periods, Liberia’s tropical moist climate allows for

trees to grow quickly in the absence of cultivation, and thus many areas that are under agricultural fallow

are classified as forest by the remote sensing technology used to produce the spatial datasets in this study.

As such, plots whose canopy cover met the official forest definition of meeting 30% canopy cover

threshold and height at maturity threshold of 5 meters were erroneously considered by field crews to be

agriculture due to evidence of previous agricultural activity. This critical distinction between land cover

and land use may not have been fully understood by the field crews, and thus results reflected a bias

toward classifying land cover type according to land use, rather than the correct land cover classification.

This bias may have been further compounded by the fact that sites visited were closer to more accessible

areas (along roads and other infrastructure), which are more likely to be subject to land cultivation.

For perhaps these same reasons, the number of points given as Forest >80% seemed very low compared

to the G/M map that had been used to determine plot locations. Upon examination of the photographs

taken, for 9 plots classified as ‘forest 30-80% canopy cover’, it appeared that they may instead be very

dense, intact forest, with >80% canopy cover. The under-sampling of this forest class could also be partly

explained by the fact that denser forests are found far from roads, and thus were less accessible by field

crew.

Reclassification

There were further difficulties related to the 4 plots that were listed as ‘swamp’ as the land use. Swamp

is not a class in the G/M classification, and conversations with the field team and further examination of

the photographs resulted in the reallocation of these plots to other classes where appropriate.

In an attempt to correct for the errors in field sampling, Ed Mitchard performed a secondary classification

largely based on the photographs collected at each plot. Plots were re-allocated to one of four classes

(Table A2):



77

Table B2. Ground truth points by re-analysed landcover type

Landcover class Number of points

Forest (>80% canopy cover) 10

Forest (30-80% canopy cover) 26

Non-forest with trees (1-30%) 23

Non-forest without trees 11

Total 70

The distribution of points between these three classes was reasonably balanced, with an under-sampling

of >80% forest expected due to the access difficulties with reaching these areas from the road (as

mentioned above).

For the purpose of the RL/REL development, the distinction between the two non-forest classes is not

especially important. Given how fast land changes between these two classes in this area, with the fast

encroachment of trees in abandoned agricultural (fallow) areas, and the generally irregular cycles of land

clearing for agriculture, errors between these two land use classifications were expected given the ~18-

month gap between the average date of the map and the field data. Therefore, the latter two classes were

combined for the accuracy analysis, leaving just three classes to be assessed: forest >80%, forest 30-80%,

and non-forest.

This fast rate of land use change was exemplified in two plots visited, which were ultimately excluded from

further analysis, as from the photos it was obvious they had been recently cleared (see photos below for

burned stumps). While the G/M map had recorded the areas as >80% canopy cover, it was clear these

areas had been recently cleared. These two plots are not considered in further analysis, so the sample size

falls to just 68 plots.



78

Plot 558, photo 7871 Plot 68, photo 94

Confusion matrix

The basic comparison between the field assessment data and the G/M map41 are reported in a Confusion

Matrix below. Confusion matrices present the full results, with the rows showing many points started as

each class in the field assessment plots and how many were classified as each class in the G/M map. The

columns show how many plots were classified as each class in the G/M map, and where they were

classified in the field assessment dataset. It can be seen that in general the map put the points in the

correct class, but with a few exceptions.

Confusion matrix

G/M map

non-forest

forest <80% canopy

forest >80% canopy SUM

Fie

ld

asse

ssm

en

t non-forest 30 2 0 32

forest <80% 2 21 3 26

forest >80% 0 0 10 10

SUM 32 23 13 68

41 The version of the G/M map used here is a derivative product at 30 m resolution, with forest areas <1 ha removed, but in the set of points chosen only 2 plots have a different value between the two maps.



79

Confusion matrices do not offer percentage accuracies, but simple calculations can provide estimates of

Producer’s and User’s accuracies. Producer’s Accuracy refers to the probability that a ground truth point

given a particular class on the ground is that class in the map, whereas User’s Accuracy is the probability

that a given pixel of that class chosen on the map is really that class in the ground.

The simplest way to explain the difference is to look at an extreme case: if our map predicted the whole

area was ‘Forest >80%’, then the Producer’s Accuracy for that class would be 100%, but for the other two

classes 0%; by contrast the User’s Accuracy for the ‘Forest >80%’ class would be the landcover proportion

of that class in the dataset. It is important to note, however, that these estimates likely do not reflect true

accuracy values as the sample size was small and not distributed across the map.

For the purposes of mapping assessments both are important, but for the purpose of this study (estimating

the extent of one class) the User’s Accuracy is the most useful. Nevertheless, the values for Producer’s and

User’s accuracy were not found to be vastly different, but it is clear the G/M map predicts too much Forest

>80%, resulting in a 100% Producer’s Accuracy for this class but a lower User’s Accuracy.

Both are given in the tables below:

Producer's Accuracy

Correctly classed

Out of total Prop correct

Field assessment non-forest 30 32 93.75%

Forest <80% 21 26 80.77%

Forest >80% 10 10 100.00%

User's Accuracy

GM correct Out of total Prop correct

Field assessment non-forest 30 32 93.75%

Forest <80% 21 23 91.30%

Forest >80% 10 13 76.92%

We can also calculate an Overall Accuracy, which is the proportion of pixels calculated correctly: 92.6 %.



80

Area-weighted calculations

While the confusion matrix offers useful results, they are not area weighted. For example, the >80% forest

class represents a far smaller proportion of the area sampled than it represents within all of Liberia (15%

of the field plot observations, 45% of the country). Nevertheless, methods exist to estimate confidence

intervals from such unbalanced ground truth datasets3.

While technically there were not enough field assessment plots to allow for the development of formal

confidence intervals on the areas of classes in the G/M map (ideally we would need 75-100 plots per class,

randomly placed within them over the whole country42), these methods were nevertheless applied, as

recommended by GFOI, to estimate the 95% confidence intervals. The results are summarised in the table

below:

Class Area in G/M

map

Unbiased area estimate,

based on field

assessment (ha)

95% C.I. (ha) Minimum

(ha)

Maximum

(ha)

non-forest 3,046,571 3,043,174 362,656 2,680,518 3,405,830

forest <80% 2,150,657 3,163,868 1,104,391 2,059,478 4,268,259

forest >80% 4,375,862 3,366,048 1,043,149 2,322,899 4,409,197

Total 9,573,090 9,573,090

The calculated confidence intervals are very wide, especially for the forest class, and are likely

overestimates. Yet, due to the small sample size, it is not possible to make that claim conclusively. What

is clear is that the unbiased estimator made possible through these methods predicts that in reality there

is more forest <80% and less forest >80% than predicted in the G/M map.

It should be emphasized that these estimates would change markedly (by hundreds of thousands of

hectares) were a single point to be misplaced or misclassified, and thus these estimates should be

considered with care. Nevertheless, they do represent the logical conclusion of this analysis based on the

field assessment data, which did show a tendency to incorrectly classify <80% cover forest as >80%.

42 Olofsson, P et al. 2014. Good practices for estimating area and assessing accuracy of land change. Remote Sensing of Environment, 148, 42-57.



81

APPENDIX C: CREATION OF LAND COVER CHANGE PRODUCTS This appendix provides more detailed information on the development of estimates of land cover change,

also addressed in Section 3 of the main report. In order to estimate activity data and emission factors it is

necessary to know the starting state of a forested area prior to it being deforested or degraded. The 2014

Metria/GeoVille has limitations for estimating RL/REL values due to these timing considerations. Thus we

need to be able to back-date the Metria/GeoVille map to a point in time before the start of the reference

period (i.e. for 2004 or earlier).

A global 30-meter resolution map of Percent Forest Canopy Cover for the year 2000 has been produced

by Hansen et al. (2013) and can be used to produce the necessary historical forest stratification. The

Hansen Percent Canopy Cover Map precisely matches an Annual Forest Loss product produced from 2000

onwards at a 30 m resolution (Hansen et al 2013), which can be used to estimate activity data. This product

analyses all available Landsat imagery and combines training data from across the planet to estimate

forest loss from 2000 to 2014. This provides the highest available resolution of data available to identify

forest loss on an annual basis. These data are freely available and annually updated, and increasingly

trusted by the forest monitoring community. The use of such datasets is recommended in the Global

Forest Observations Initiative’s Methods and Guidance Document “Integrating Remote-sensing and

Ground-based Observations for Estimation of Emissions and Removals of Greenhouse Gases in Forests”43

(referred to as ‘MGD’), and specifically encouraged in the recently published ‘Module 2’ of the MGD.

The only other systematically produced products giving deforestation are reliant on MODIS data, with a

best resolution of 250 m. This was coarser than required in this landscape, where small patches of

deforestation dominate, and thus would be likely to underestimate total change. It would separately have

been possible to manually classify Landsat scenes as part of the project, producing our own landcover and

landcover change products from 2000 onwards. However, a lack of ground data and extensive cloud cover

meant that there was no reason that maps produced through the project would be any more accurate

than the ready produced Hansen et al. (2013) global deforestation maps, and certainly there would be no

means to assess their relative accuracy. In fact, it would have been likely that any maps we produced

would have had lower accuracy as we would have considered fewer Landsat scenes: the Hansen et al.

products use all Landsat data collecting, involving thousands of scenes over Liberia, whereas we would

have been limited to using at most tens of scenes. Using fewer scenes increases the proportion of the

43 http://www.gfoi.org/methods-guidance/




82

country that would not be considered due to cloud cover at any one point, and increases errors by not

fully considering time series data.

The Hansen et al canopy cover map was used to create estimates of activity data, however, prior to its use

a local correction was made, as also recommended by the MGD. This Appendix explains the correction

made for this data series to best reflect the land cover realities in Liberia.

Prior to this correction, the global Hansen et al. canopy cover map did not match up with the

GeoVille/Metria map – it underestimates the area of >80% cover forest, and over-estimates the area of

30-80% canopy forest, with almost no areas of the country given an area <30% cover. This is not surprising

as the Hansen et al. canopy cover product relies on a global algorithm, and Liberia, with its dense grass

and crop cover and quick-growing fallow, is considerably greener for a given canopy cover percentage

than most of the rest of the world. However, this discrepancy requires that an adjustment needs to be

applied to use it to produce reliable classes in the past.

As there are no independent reference data available for 2000, nor any independent high resolution

datasets, a visual comparison with the Metria/GeoVille classification for areas without deforestation was

performed. This relied on the theory that areas without deforestation should have the same distribution

of classes in both 2000 and 2014. Thresholds were found that appeared to replicate the Metria/GeoVille

classes to a good extent in areas where deforestation had not occurred. The thresholds are shown in Table

C1, displayed in C1, and the resulting forest areas given in Table C2. Figure C1 includes a map of

deforestation density within 1 km pixels. This is included as the differences between canopy cover in 2000

and 2014 should, if the correction has been performed appropriately, be explained by this change map.



83

Figure C1. Maps of three forest classes with ‘corrected’ Hansen et al. thresholds



84

Table C1. Canopy cover thresholds for three forest classes, Metria/GeoVille (2014) & Hansen (2000)

Forest Class GeoVille range

Hansen range

Dense forest (> 80 %) >80 >72

Secondary forest (30-80 %) 30-80 56-71

Trees in agricultural mosaic (10-30 %) 10-30 47-55

Table C2. Areas of three forest classes, Metria/GeoVille (2014) & Hansen (2000 – corrected)

Forest Class GeoVille Hansen (corrected)

Difference in area

canopy cover ha, 2014 ha, 2000 ha, 2000-2014, integrating deforestation, degradation and regrowth

>80 % 4,583,778 5,778,415 -1,194,637

30 - 80 % 2,188,842 2,485,622 -296,780

1-30 % 1,462,931 953,070 509,861

Total 8,235,551 9,217,107 -981,557

Total >30% 6,772,620 8,264,037 -1,491,417

Accuracy of Hansen et al. product

The Hansen et al. (2013) deforestation product was published in the respected journal Science, with an

accompanying website distributing the raw data44 receiving hundreds of thousands of download requests

over the subsequent months. Further, a website run by the World Resources Institute funded by

governments, NGO’s and the UN called Global Forest Watch allowed the querying and display of the data45,

receiving millions of views as well as widespread media coverage. The attention and widespread use of

the data led to many attempts to validate its accuracy.

The original paper contained a rigorous validation exercise. Different biomes were treated separately; for

the tropics, the focus here, 628 x 120m x 120m sample blocks were chosen, distributed between no change,

gain and loss pixels. Local experts used time series of very high resolution independent remote sensing

data to assess the proportion of forest loss or gain through the time series, and then these were compared

44 https://earthenginepartners.appspot.com/science-2013-global-forest 45 http://www.globalforestwatch.org/

https://earthenginepartners.appspot.com/science-2013-global-forest

http://www.globalforestwatch.org/



85

to the original results. The test data were not used in generating the product, and so were independent.

For the tropics the results suggested the product was unbiased (mean difference in loss per block is 0, with

a Standard Error of ±0.5 %), and reasonably accurate, with a User’s Accuracy of 87% for ‘Loss’ and 99.8%

for ‘No change’.

Uncertainty of forest loss data

A lack of historical field data made a Liberia-specific assessment of the accuracy of the Hansen et al. forest

loss data impossible. Only the 2014 landcover map was validated by the field data collection in 2016, and

no suitable historical data was found. This is contrary to guidance in the MDG, which suggests that global

deforestation datasets should be compared to a large number of points, stratified randomly across the

country, derived from either field data or high resolution remote sensing data.

An attempt was made to use L-band radar data available from 2007-10 to perform validation, but a lack

of local biomass plot data made calibration of the data layers impossible. Without biomass plot data for

calibration, it was not possible to ascertain whether a change in radar backscatter was sufficient to be

deforestation or degradation, nor to set thresholds for forest/non-forest. Ultimately validation can only

proceed if there is greater confidence in the test dataset (the L-band radar data) than the dataset to be

validated: this was not possible in this case. Such validation could be performed going forward, if biomass

data are acquired.

Similarly, an initial examination of available archives of high resolution optical data suggested no suitable

time series of sufficiently high resolution imagery (<1.5 m) was available over the study period. At least 3

cloud-free images, each separated by at least 12 months, would have been necessary, and no suitable

areas were detected after examining GeoEye, Worldview and SPOT archives. This situation has improved

markedly since 2014 with many new satellites launched and an increased interest in collecting data over

forests, but more recent high resolution data cannot assist with assessing the accuracy of historical change

data.

Therefore we must rely on extensive independent ground truthing exercise performed by Hansen et al.

(2013). In the Tropical climate domain they used local experts examining very high resolution imagery to

assess 628 reference areas. These had an overall accuracy of 99.5 %, with a 95% confidence around that

accuracy percentage of 0.2. However, the accuracy was much higher for the ‘no loss’ class (99.7 ±0.1 %)

than the Loss class (87.0 ±4.7 %). The full error matrix is given in Table C3 below.



86

Table C3. Loss Error Matrix expressed as a percent of area. Data taken from Hansen et al. (2013), based on 628 observations

over the tropical climate domain.

Reference (%)

Loss No Loss Total

Map (%) Loss 1.50 0.22 1.72

No Loss 0.30 97.98 98.28

Total 1.80 98.2

Applying the Olofsson et al. (2013)46 method this allowed an unbiased area estimate for the Reference

Level, based on pantropical accuracy assessment data, divided by strata assuming an equal chance of error

between the 30-80 and >80 % strata. This is shown in Table C4 for the whole reference period, but can

also be applied evenly to every year.

Table C4. 95% confidence ranges and unbiased area estimates based on Hansen et al. (2013) ground truth data for entire

reference period (2005-2014).

Class Original area estimate (ha)

Unbiased area estimate (ha)

95 % Confidence interval (ha)

Minimum (ha)

Maximum (ha)

Unchanged forest

6,568,973 6,549,932 128,499 6,421,433 6,678,431

Deforested, forest 30-80%

97,472 103,605 46,692 56,914 150,297

Deforested, forest >80%

205,135 218,043 64,248 153,795 282,291

Total 6,871,580 6,871,580

These can also be used to give adjusted means and confidence ranges for the deforestation rate over the

period, as summarised in Table C5 below.

46 Olofsson, P., Foody, G.M., Stehman, S.V., & Woodcock, C.E. (2013). Making better use of accuracy data in land change studies: estimating accuracy and area and quantifying uncertainty using stratified estimation. Remote Sensing of Environment 129:122-131



87

Table C5. Annual mean deforestation rates with 95% confidence intervals.

Class Mean estimate of deforestation rate

95% Confidence interval on deforestation rate

Confidence interval proportion of total

Deforestation rate, forest 30-80%

1.137% 0.513% 45.1%

Deforested, forest >80%

0.382% 0.113% 29.5%

These give broad ranges for estimated deforestation rates. However, as these are based on a pantropical

accuracy assessment, itself not stratified by forest type, these uncertainty ranges are themselves highly

uncertain.

Hansen et al. did not report on percentage errors of Omission and Commission in their results, which are

ideal for estimating the impact of errors on result statistics. Omission errors are where the map has missed

deforestation that occurred in reality, and commission where the map predicts deforestation occurred in

an area where in fact no change occurred. A number of studies in this area are ongoing, and are yet to be

published as the typical research and publication cycle takes a minimum of 2-3 years. However two

independent assessment are available:

1. Mitchard et al. (2015)47 used 5m resolution RapidEye data and field knowledge to test the

Hansen et al. data in the Brazilian Cerrado and Ghana’s tall forest. Two different methods of

interpreting the RapidEye data were used, giving slightly different results – manual

interpretation of changes, and semi-automated classification. The overall findings were that the

Hansen et al. data performed very well in the Brazilian Cerrado, with Commission and Omission

rates both less than 15% overall. This example is probably most comparable to the Liberia case,

with a mixture of tall forest, scrub and agriculture. In Ghana the Hansen et al. product was found

to have performed poorly, missing extensive degradation occurring in tall forest blocks: but this

is probably irrelevant to the Liberia example where the Hansen et al. data is not being used to

assess degradation.

2. Muller et al. (2016)48 assessed long term deforestation trends throughout the Brazilian Amazon,

using their own deforestation assessment as well as the Hansen et al. dataset. They could not

47 Mitchard, E., Viergever, K., Morel, V., & Tipper, R. Assessment of the accuracy of University of Maryland (Hansen et al.) Forest Loss Data in 2 ICF project areas. http://ecometrica.com/wp-content/uploads/2015/08/UMD_accuracy_assessment_website_report_Final.pdf 48 Muller, H., Griffiths, P., & Hostert, P. 2016. Long-term deforestation dynamics in the Brazilian Amazon – Uncovering historic frontier development along the Cuiba-Santarem highway. International Journal of Applied Earth Observation and Geoinformation. 44, 61-69.

http://ecometrica.com/wp-content/uploads/2015/08/UMD_accuracy_assessment_website_report_Final.pdf

http://ecometrica.com/wp-content/uploads/2015/08/UMD_accuracy_assessment_website_report_Final.pdf



88

conduct a formal accuracy assessment as their own classification only had an accuracy of 85%,

but found that their results closely matched the Hansen et al. product in annual trends and

absolute values, though with a slight tendency to underestimate total area through missing

small clearings.

These results, while not specific to Liberia, suggest that the Hansen et al. data is sufficiently accurate to

create unbiased estimates of forest change suitable for the further analyses done to estimate Reference

Levels. If such data were to be used for MRV purposes we would recommend a combined ground and

high-resolution remote sensing campaign, as will be detailed in Version 2.0 of the GFOI Methods and

Guidance Document due out this summer, to further assess the specific accuracy of the Hansen et al.

method by landcover type and size of disturbance in Liberia.



89

APPENDIX D: EXISTING BIOMASS DATA, ALLOMETRIC

EQUATIONS AND BIOMASS EXPANSION FACTORS

Existing Biomass Data, Allometric Equations and Biomass Expansion

Factors

Emission factors are measures of the emissions and removals of greenhouse gases per unit of activity data,

usually expressed in units of t CO2e ha-1. Emission factors for land use change are generally developed by

estimating biomass and carbon stocks of the relevant pools and land cover types.

Various sources of data may be used to estimate forest biomass and develop emission factors. Potential

sources for generating emission/removal factors include:

Carbon measurement inventories including ground measurement, allometric equations and

remote sensing techniques. These rely on allometric models that relate the biomass of trees

with certain measureable morphological features (e.g. diameter and height) to indirectly

quantify aboveground and belowground tree biomass estimates.

Forest or timber inventories that provide data on the number or trees per hectare or the volume

of timber. These rely on biomass expansion factors to estimate aboveground biomass.

Academic and other research studies that have previously produced biomass and/or carbon

estimates.

Existing relevant Allometric Equations

The underlying data to estimate carbon stocks are collected during field inventories. These data are then

converted to biomass estimates using allometric equations. However, before biomass data is collected a

field team must determine which allometric equation(s) it will use, so that it knows what must be

measured in the field. Live trees contain the majority of biomass in most forests, and the informed

selection and verification of allometric models to estimate biomass is a crucial step in developing accurate

estimates of forest carbon stocks49.

49 Biomass stocks are converted to carbon stocks using the IPCC default carbon fraction of 0.47.



90

There are a number of variables that are commonly used to estimate tree biomass. Most allometric

equations are developed to estimate total aboveground biomass, however, some equations are created

to estimate different components of the tree such as stem, branches, leaves etc. Variables commonly

within equations include:

Stem/trunk diameter at breast height (at 1.3 m aboveground; DBH)

Stem diameter at stump height (DSH) (common for multi-stemmed trees)

Basal area

Total height

Botanical identification

Wood density

Site quality

Tree age (common for trees grown in plantations)

Crown width (common for shrubs)

Climate (environmental stress factor) Of these variables, DBH and wood density are easiest to attain and provide the most reliable inputs (Brown

1997; Chave et al. 2005).

Developing allometric equations is a labour intensive and costly process. If an existing equation is found

to be appropriate, it is much more cost effective to use that equation, rather than developing a country-

specific equation. If possible, it is generally recommended to verify existing allometric equations by

destructively sampling a small number of trees to directly measure biomass.

Equations have been developed across Africa and globally across the tropics. Tables D1 & D2 provide existing biomass allometric equations, including Africa specific equations (Table D1) and pan-tropical equations (Table D2). The following variables are used in these equations:

AGB = aboveground biomass (kg) D = diameter at breast height, at 1.3 m aboveground (cm) BA = basal area (cm2) H = height (m) ρ = wood density (g/cm3) E = environmental stress factor (unitless)



91

Table D1: Select Region/Country specific allometric biomass equations for Africa

Area/ Country

Equation Forest Type n R2 Tree Size Source

Central Africa, Congo Basin, Cameroon

AGB = ρ * exp (-1.183 + 1.940 × In(D) + 0.239 × (In(D))2 – 0.0285 × (In(D))3)

Lowland tropical forest (moist forests)

138 0.988 RSE= 0.188

NA Fayolle et al. 2013

Cameroon

ln(AGB) = -2.1801 + 2.5634 x ln (D)

Moist tropical forest

443 0.9671 RSE=0.444

1-148cm DBH

Djomo et al. 2010

ln(AGB) = -3.2249 + 0.9885 x ln (D2 x H)

274 0.971 RSE=0.437

1-138cm DBH

ln(AGB) = -2.4733 + 0.2893 x (ln (D))2 – 0.0372 x (ln(D))3 + 0.7415 x ln (D2 x H) + 0.2843 x ln(ρ)

274 0.9717 RSE=0.437

1-138cm DBH

Cameroon

AGB = -3.37 + (0.02483 X D2H) Regenerating tropical forest species

14 0.99 5-120cm DBH Deans et al. 1996 AGB = -30.87 + 0.7684 * BA

Ghana AGB = 0.30 * D2.31

Tropical Rainforest, Wet evergreen forests

42 0.93 2-180cm DBH Henry et al. 2010

Central Africa

ABG=exp(-4.0596+4.0624×ln(D)-.0228 x (ln(D))2 + 1.4307 x ln(ρ)) Evergreen

rainforest mixed semi-deciduous species

101 0.944

11.8-109.4cm DBH

Ngomanda et al. 2014

ABG=exp(-2.5680 + 0.9517 × ln(D²×H) + 1.1891 x ln(ρ))

AGB = 0.1083 x (D2 x H)0.0138 30 0.99

AGB = 0.0558 x H0.0113 30 0.97



92

Table D2. Pan-tropical allometric biomass equations

Life Zone Equation n R2 (adj)

RSE50 Tree size

Source

Dry AGB = 34.4703 - 8.0671 * D + 0.6589 * D2

32 0.67 0.02208 58-39 cm DBH

Brown et al. 1989

Moist

AGB = 38.4908 – 11.7883 * D + 1.1926 * D2

168 0.78 0.06181 5-130 cm DBH

AGB = exp(-3.1141 + 0.9719 * ln(D2 * H))

168 0.97 0.1161

AGB = exp(-2.4090 + 0.9522 * ln(D2 * H * ρ))

94 0.99 0.06079

H = exp(1.0710 + 0.5677 * ln(D))

3824 0.61 0.07495

Wet

AGB = 13.2579 – 4.8945 * D + 0.6713 * D2

69 0.90 0.02247 5-110 cm DBH

AGB = exp(-3.3012 + 0.9439 * ln(D2 * H))

69 0.90 0.2110

H = exp(1.2017 + 0.5627 * ln(D))

69 0.74 0.4299

Dry

AGB = exp(-1.996 + 2.32 x ln(D))

28 0.89 5-40cm DBH

Brown 1997 AGB = 10(-0.535 + log10(BA)) 191 0.94

5-30cm DBH

Moist51 AGB = exp(-2.289 + 2.649 x ln(D) - 0.021 x ln(D)2)

226 0.98 5-148cm DBH

Wet AGB = 21.297 – 6.953 x D + 0.740 x D2

169 0.92 4-112cm DBH

Dry

AGB = exp(-2.187 + 0.916 * ln(ρ * D2 * H))

316 0.99 0.311

5-63.4cm DBH Chave et al.

2005

AGB = ρ * exp(-0.667 + 1.784 * ln(D) + 0.207 * (ln(D))2 – 0.0281 * (ln(D))3)

316 0.99 0.356

Moist AGB = exp(-2.977 + ln(ρ * D2 * H))

1349 0.99 0.311 5-138cm DBH

50 All Brown et al. (1989) values listed in this column for are MSE, not RSE 51 The moist equation is updated from Brown 1997 with additional destructive sampling data and a new form of the equation



93

Life Zone Equation n R2 (adj)

RSE50 Tree size

Source


1349 0.99 0.356

Wet

AGB = exp(-2.557 + 0.940 * ln(ρ * D2 * H))

143 0.99 0.311

5-133cm DBH


143 0.99 0.356

Pantropical

ln(AGB) = -1.8222 + 2.3370 x ln(D) + 0.1632 x (ln(D))2 - 0.0248 x (ln(D))3 + 0.9792 x ln(ρ)

1816 0.973 0.3595 >10cm DBH Feldpausch et

al. 2012 ln(AGB) = -2.9205 + 0.9894 x ln(D2 x ρ x H)

1816 0.978 0.3222 >10cm DB

Pantropical

AGB = 0.0673 * (ρ * D2 * H)0.976

4004 0.357 5-180cm DBH

Chave et al. 2014

AGB = exp(-1.803 – 0.976 * E + 0.976 * ln(ρ) + 2.673 * ln(D) – 0.0299 * (ln(D))2)

4004 0.431

With a multitude of varying options for calculating AGB, a pan-tropical allometric equation, such as Chave

et al. (2005, 2014), is extremely useful. The validity of the Chave et al (2005) equation has been confirmed

in studies across Africa, areas where uncertainty in the 2005 models was thought to be due to climatic

variations. Chave et al. (2014) published an improved allometric equation inclusive of a variable

representative of climatic effects on tree growth. In many cases, these pan-tropical equations have been

shown to be more reliable than country-specific equations, primarily due to increase sample size (Goslee

et al, 2015). Therefore, the Chave el at 2014 equation is recommended for use in Liberia, in association

with relevant inventory data.

Use of biomass expansion factors

Allometric equations are one way to use measurement to estimate above ground tree biomass and therefore carbon, but tree biomass can also be estimated from volume over bark of merchantable growing stock wood (VOB) by “expanding” this value to take into account the biomass of the other aboveground components – this is referred to as the biomass expansion factor (BEF) (Walker et al. 2013).



94

When data on tree volumes exist the BEF can be used to estimate a trees biomass and therefore carbon (Goslee et al. 2015). Above ground biomass can be estimated base on existing volume per ha data. The primary data needed for this approach is VOB/ha and a volume-weighted average wood density (oven dry mass per unit of green volume in t/m3). Biomass density can be calculated from VOB/ha by first estimating the biomass of the inventoried volume

and then "expanding" this value to take into account the biomass of the other aboveground components

as follows:

Aboveground biomass density (t/ha) = VOB ∗ WD ∗ BEF

Where:

VOB = volume over bark of free bole from stump or buttress to the crown point or first main branch (m3/ha)

WD = volume-weighted average wood density (t/m3) BEF = biomass expansion factor (ratio of aboveground oven-dry biomass of trees to oven-dry

biomass of inventoried volume, unitless)

The IPCC (2006) report provides a method for using VOB to estimate the AGB of forests—in this report it

refers to the Biomass Conversion and Expansion Factor (BCEF) that is the product of the BEF and wood

density and values of BCEF are given for a range of VOB classes. The values for tropical humid natural

forests range from 9.0 (range 4-12) for a VOB of <10 m3/ha to 0.95 (range 0.7-1.1) at VOB >200 m3/ha

(Table 4.5 in Vol. 4, Ch. 4 of IPCC 2006).

Existing relevant Biomass Data

Once an allometric equation is selected field measurements can commence. Traditional ground-based

forest inventories are based on statistical sampling, where field data of easily measurable tree parameters,

such as diameter at breast height (DBH measured at 1.3 m from the ground) are collected. Collecting

carbon stock information for several samples (plots) across the population of interest generates summary

statistics about the population, such as the mean (average) carbon stock, and the measured variation

among samples. Tables D3 and D4 below show carbon stock data for two specific regions in Sierra Leone

and Guinea Bissau that could relate to carbon stock in Liberia.



95

Table D3. Above and below ground tree carbon stocks (CAB_Tree,I and CBB_Tree,i, respectively) and soil carbon stocks (CSOC,i) for the

Gola Rainforest National Park in Sierra Leone in t CO2e ha-1 (Netzer and Walker, 2013).

Carbon Pool

Strata 1 (GRNP Central/North) Strata 2 (GRNP South)

Number of Plots

Mean Stock

95% CI 95% CI as % of mean

Number of Plots

Mean Stock


t CO2 ha-1 t CO2 ha-1

CAB_Tree,i 353 654.7 48.4 7% 49 582.5 76.6 13%

CBB_Tree,i

157.1 11.6 7%

139.8 18.4 13%

CSOC,i 18 253.9 30.6 12% 29 192.3 24.4 13%

Both strata in the Gola Rainforest National Park (GRNP) in Sierra Leone can be classified as moist evergreen

forest. The results of this extensive survey work showed that the forests across the GRNP were relatively

homogenous in species composition (same forest type), however there were significant differences in

carbon stocks between Gola South, and Central/North. It was hypothesized that the difference between

the stocks in the 2 areas was due to past management histories, the southern block having been more

extensively logged than the central or northern blocks, thus resulting in a forest with lower carbon stocks

but with potential for significant re-growth.

Table D4. Above (CAB_Tree,i) and below (CBB_Tree,i) ground tree carbon stock) for closed forest in Canthanhez Park in Guinea Bissau

in t CO2e ha-1 (Amaro et al., 2012).

Carbon Pool Number of Plots

Mean Stock


t CO2 ha-1

CAB_Tree,i 45 320.3 15.26 17%

CBB_Tree,i 45 85.9 4.53 19%

The Cantanhez National Park in Guinea Bissau, with an area of 106,500 hectares is located in the

administrative region of Tombali, covering the Bedanda sector in Guinea Bissau. Cantanhez forests

represent the latest patches of sub-humid forest which form part of a larger area that extends to the

south, into Guinea Conakry. Terrestrial vegetation in the Cantanhez Park consists of patches of dense

mature forest in a mosaic of patches of secondary forests from cultivation and fallow by shifting

agricultural practices. Mangroves cover a large proportion of the area of the park, particularly to the south

and western regions, in the margins of the Cumbijã River (Amaro et al., 2012).



96

In addition to these data from nearby regions, there have been some studies estimating carbon stocks

within Liberia. In 2005 and 2006, a forest inventory was conducted, with 405 sampling clusters located

across the country, covering five different land cover classes, including three forest classes: agriculture

degraded forest, open dense forest, and closed dense forest. However, due to accessibility and time

constraints, data were collected on only 167 clusters, 127 of which were on forest classes (Hess and

Trainer, 2006). The carbon values estimated based on this inventory data are provided in Table D5,

however, the inventory did not meet its minimum criteria for a standard error ±10% (Ebeling and Asare,

2011). Liberia’s R-PP states that error estimates from this inventory were too high to allow its use to

develop a database of biomass stocks.

Table D5. Carbon estimates based on Liberia’s National Inventory (adapted from Ebeling and Asare, 2011).

Forest type Number of clusters

AGB BGB Total

t CO2 ha-1

Agriculture degraded forest

18 400 81 480

Open Dense Forest 42 598 114 711

Closed Dense Forest 67 631 121 752

Additionally, in 2011, an initial carbon stock assessment and capacity building exercise was conducted on

the Wonegizi REDD+ project in Lofa County. During this exercise, three one hectare plots were established,

one in each of three land cover types: forest, coffee, and cocoa. From the forest plot, aboveground carbon

stocks were estimated at 108 t C/ha (Asante and Jengre, 2010). A more complete inventory was conducted

on the Wonegizi project in 2013. In this effort, carbon stocks were established for primary forests (718.7

t CO2e) and secondary forests (392.3 t CO2e), based on a total of 37 plots (FFI and RSS GmbH, 2014).

Gaps in Existing Data

Although some carbon stock data exist in the region as described above, there are no existing data in

Liberia that can be used to develop reliable country-specific emission factors or to assess whether existing

data correlate closely enough to Liberia forest carbon stocks. To develop these, therefore, it would be

necessary for Liberia to develop a forest sampling scheme and undertake a forest inventory (an annex

providing further guidance on this will be provided in the final report). It is recommended that such an

inventory be conducted when time and resources allow, in order to develop an estimate of carbon stocks

and emission factors with an acceptable uncertainty level.



97

However, because this is outside the scope of this current effort, we must rely on existing datasets from

outside Liberia to develop emission factors. These datasets can be used to develop Tier 1 emissions

factors, because while they are not based on IPCC defaults, they are also not derived from country-specific

data, as required to Tier 2 emission factors. Recommendations for the capacity and resources needed to

develop Tier 2 emission factors will be provided in the final capacity building strategy and the final report.

Global Biomass Datasets

There have been two widely publicised maps of pantropical aboveground biomass: Saatchi et al. (2011)

which is at a 1 km resolution, and Baccini et al. (2012) at a 500 m resolution. Both are produced using

broadly similar methods, extrapolating rare forest height measurements from a LiDAR satellite that

operated in the mid-2000’s (ICESat GLAS) using other remote sensing data (mostly MODIS). The height

estimates from GLAS are in both cases converted to aboveground forest biomass using a few hundred field

plots – the actual plots and allometric equations used do differ between the studies. Similarly they differ

by date – the Saatchi et al. map is for the early 2000’s, Baccini et al for 2007.

The two maps have been compared to each other, finding significant differences (Mitchard et al. 2013),

and compared to field data finding some potentially large regional biases (Mitchard et al. 2014), but are

thought useful as a first estimate of aboveground biomass distribution in a region. It was thought

interesting to compare these two for Liberia, and use them to find an approximate biomass value stored

within the different forest classes identified.

Recently, a ‘consensus’ map has been produced using inputs from these two pantropical maps and a

dataset of field plots and local high resolution maps covering almost 15,000 1 km pixels (Avitabile et al.

2015). This map is expected to have lower regional biases, and contains plots from within Liberia itself as

well as various other countries in West Africa. This dataset is also included in the comparison.

Maps

The maps (Figure D1) produce strikingly different distributions of biomass across the country: the figure

below shows the three maps displayed with an identical scale.



98

Baccini et al. Saatchi et al. Avitabile et al

Figure D1. Maps showing distribution of forest biomass in Liberia, based on available global datasets.

In particular the Baccini et al map predicts high AGB only in the north of the country, whereas the Saatchi

et al. and Avitabile et al. maps predict far higher biomass in the north and south of the country, and to the

west. There is also more contrast between high and low biomass areas in the Saatchi et al. than Baccini et

al. map, and more still in the Avitabile et al. map.

Total Carbon Stocks

The mean biomass and total carbon stock predicted by the Saatchi et al. and Baccini et al. maps differ, but

not by as much as might have been thought looking at the maps (Table D6). The lower AGB values in

agricultural areas in the Saatchi map cancel out its higher values in the forested areas to some extent. The

Avitabile et al map on the other hand predicts considerably higher AGB – it has a larger area of high

biomass forest than the other two, but also has far higher values, going up to 567 Mg ha-1 for its highest

pixel, compared to 423 Mg ha-1 in Saatchi et al. and 435 Mg ha-1 in Baccini et al.



99

Table D6. Biomass values across Liberia based on three global datasets. (Carbon is calculated by multiplying biomass by 0.47.

It should be noted that both forest and non-forest land is included in the ‘per ha’ figures, which is appropriate as these are 1

km maps, but means that these numbers will be an underestimate compared to forest area alone.)

Product Mean biomass per ha

Baccini et al. (2012) 188.7 Mg ha-1

Saatchi et al. (2011) 209.7 Mg ha-1

Avitabile et al. (2015) 244.9 Mg ha-1

Mean aboveground biomass density by vegetation type

The three maps were queried to calculate the mean carbon stock in the three forest classes identified by

Metria/GeoVille in their 2014 landcover map of Liberia at a 5 m resolution.

Table D7. Carbon stocks in Liberia by forest class, based on global datasets, shown in CO2e ha-1

Forest Class Baccini et al. Saatchi et al. Avitabile et al.

t CO2 ha-1

Forest >80% 364 436 566

Forest 30-80% 317 333 365

Forest <30% 291 311 302

The biomass values increase in step between the three maps, as would be expected given their overall

biomass numbers. However, of particular interest is the increased contrast between vegetation types from

Baccini to Saatchi, then Saatchi to Avitabile. The highest canopy cover forest class is only 25 % bigger than

the lowest cover in the Baccini map, but 87 % bigger in the Avitabile map. We would expect a big difference

in stocks between these classes, so this suggests the Avitabile map could be the most reliable. In addition,

the Avitabile carbon stocks are based on actual field data collected in Liberia (albeit limited) and used to

locally weight and average the Baccini and Saatchi maps. Further, the overall carbon stocks are more in

line with the values that have been collected in Liberia (see Table D7). For this reason, it is expected that

the Avitabile data provide a more accurate estimate of carbon stocks in Liberia. Nonetheless, we have



100

used the Baccini estimates to develop provisional emission factors, as they are more conservative, and

therefore less likely to overestimate historical emissions. Regardless, this constitutes only Tier 1 emissions

factors, which may not be accepted by FCPF or UNFCCC. It is therefore highly advisable for Liberia to

develop country-specific emissions factors.

APPENDIX E: RECOMMENDATIONS FOR IMPROVING

DEFORESTATION ACTIVITY DATA FOR LIBERIA

Assessing deforestation rates in the past for Liberia is limited by two factors: the availability of

satellite data and the availability of field data. These factors are far less limiting in the future, with

satellite data provision (for both unrestricted and commercial data) far more plentiful from 2014-

15 onwards. There is therefore more scope to improve mapping of deforestation for MRV

purposes than for improving past Activity Data, but both can be improved from current estimates.

Here are provided 1) options for developing activity data under the future MRV, and 2) potential

methods for improving estimates of historical emissions.

Recommendations for historical Activity Data mapping

Improving the current maps of past activity data is limited by data availability, in terms of both

satellite data and field data. While the methods described in this report are appropriate for

developing Approach 3 activity data, they could be improved if time and resources allow. The

methods described here could improve the accuracy of forest change and activity data estimates,

but would require significant investment to undertake. Therefore, Liberia must decide if such

improvement are required and/or desired.

The recommended method below (Box 4) relies on a combination of the purchase of high

resolution optical data and analysis of mosaics of free Landsat data. The accuracy would likely be

higher than using the Hansen et al. dataset, not because the input data quality would be higher,

but because the forest/non-forest definition and forest classifications would be tuned to Liberia.

If high resolution and Landsat data approach is considered unfeasible due to a lack of cloud free

data, an alternative method could be considered, still using high resolution optical data for

training/testing, but using radar data rather than Landsat data for the classification. However,



101

radar data is only available from 2007 onwards, which would not span the range of the Reference

Period, and thus would not be an ideal choice.



102

Overview of Historical Land Cover Map Creation Option

Below is an overview of the steps required to develop a series of land cover maps as an

alternative to using the Hansen et al database. This is not required and involves significant

financial and human resources.

1. Mapping past deforestation

1.1. High resolution reference data order and download

1.1.1. Reference data is needed to calibrate and validate the past forest/non-forest

maps. As no suitable historical field data exists, we recommend the use of

hyperspatial remote sensing data with a resolution smaller than 2.5 m per

panchromatic pixel, so individual trees can be seen.

1.1.2. Such data is available from three satellites earlier in the time period: the

IKONOS (launched 2000, 1 m resolution) QuickBird satellite (launched 2001,

80 cm resolution) and SPOT 5 (launched 2002, 2.5 m resolution). The range of

satellites increases from 2007 with the launch of WorldView satellites and

further SPOT satellites.

1.1.3. Data availability over Liberia is not very high, due to cloud cover and few

commercial clients ordering data for forest monitoring in the mid-2000’s. We

therefore recommend an opportunistic approach to finding data stacks,

involving archive searches on all commercial satellite archives with sufficient

resolution throughout the period.

1.1.4. Ideally a set of high resolution training areas (at least 3), each at least 500

km2 in size, would be found, with high resolution data stacks at least 3 deep

available. For example, ideally three contrasting areas could be found with

high resolution observations c. 2003, 2007, 2010 and 2015.

1.2. Landsat data preparation

1.2.1. Landsat 5, 7 and 8 data should be downloaded covering all of Liberia, with

years matching those in the training data series. At least four dates should be

used, for example 2005, 2008, 2011 and 2014.

1.2.2. For each year, cloud-free scenes from as close as possible at date to each

other should be mosaicked. Any remnant clouds should be filled in by other

scenes from the same year, or where necessary, pixels from the year before

or after. Colour balancing should be performed to match the spectral

characteristics of the mosaics, using for example the features in the Semi-



103

Automated Classification Plugin in QGIS (open source), or the in built

mosaicking tool in ENVI (commercial software).

1.2.3. For 2015, the Landsat 8 mosaic already prepared by Geoville/Metria could

potentially be used.

1.3. Creating training data from high resolution optical data

1.3.1. Classifying hyperspectral remote sensing data can produce low accuracy

products, as shadows, texture, different look angles and phenology can all

lead to confusion in automated classifiers. By contrast, a trained human

interpreter can normally assess whether a given pixel is over a tree or bare

ground with high accuracy. Therefore the data will be analysed by creating

virtual plots on the high resolution data and giving % canopy cover values to 1

ha square regions of data.

1.3.2. Using QGIS (or other GIS software) 1000 random points should be generated

in each of the three areas where hyperspectral image stacks are available.

Around each point, a 100 x 100 m square plot with a 1m grid of points should

be created.

1.3.3. A team of interpreters should classify each point within a plot as either ‘tree’

or ‘not-tree’. It is recommended that each plot is classified at least twice,

blind, by different interpreters, in order to confirm that results are accurate.

If the different classifications agree on a total canopy cover for the plot within

10%, the result should be averaged; if they differ by more than 10 % a third

interpretation should be performed, or the plot removed.

1.3.4. The grid of points is converted to a canopy cover % for that hectare, resulting

in 3000 training points spread across the country.

1.4. Classification

1.4.1. The training data in 3.3 would be used to classify the Landsat data prepared

in 3.2. The most accurate results would come from the use of just two

classes, non-forest (canopy cover <30% as one class) and forest (canopy cover

>30 % as the other class). However, the forest class could, if desired, be split

into more classes to produce more useful, if less accurate, results. A cut off

for high canopy cover forest of >80%, as used by the Metria/Geoville

classification, worked reasonably well for the 2014 map.



104

1.4.2. The best classification result would probably come from the use of

proprietary software, for example the use of the Support Vector Machine or

Neural Network algorithms in ENVI52, or an object-based approach using

Ecognition53. However very good results could probably be achieved from

using open source software such as R (where RandomForest54 or other tools

could be used to perform a robust machine learning classification), or using

the Semi-Automated Classification plugin55 in QGIS.

1.4.3. The map accuracy should be tested by withholding 50 % of the training data

for testing purposes. A forest/non-forest accuracy of >90 % should be

achievable, ideally >95%. If the forest class is split into more classes, the

accuracy between these classes would be good if >85%.

1.5. Validation

1.5.1. As stated above, the classifier would be tested by holding back 50 % of points.

This will provide an accuracy assessment for the classifications only. It does

not validate the change map performed by differencing the classifications

produced in 2003 with 2007, and so on, as most of the ground truth points

will not have changed class in that period.

1.5.2. Instead therefore the change map should be validated after creation by

targeting ‘change’ pixels in the Landsat maps within the boundaries of the

three high resolution areas. 300 random change pixels and 300 random

unchanged pixels within each of three areas, for each change period, should

be assessed by eye against the high resolution data and marked as ‘changed

or unchanged’. The operator performing this validation should not be

informed whether the pixels being checked as changed or unchanged, nor

whether the change direction is positive or negative.

1.5.3. This will allow a calculation of the error and bias of the deforestation product

1.6. Estimated costs

52 http://www.exelisvis.co.uk/ProductsServices/ENVIProducts/ENVI.aspx 53 http://www.ecognition.com/ 54 https://cran.r-project.org/web/packages/randomForest/index.html 55 https://plugins.qgis.org/plugins/SemiAutomaticClassificationPlugin/

http://www.exelisvis.co.uk/ProductsServices/ENVIProducts/ENVI.aspx

http://www.ecognition.com/

https://cran.r-project.org/web/packages/randomForest/index.html

https://plugins.qgis.org/plugins/SemiAutomaticClassificationPlugin/



105

1.6.1. Data costs for archive SPOT data is approximately $3/km2 (for 2.5 m

Pansharpened colour), or for Quickbird/IKONOS around $15/km2. In both

cases discounts may be available. For the 1500 km2 x 4 periods requested

that suggests a total cost of about $18,000 for SPOT 5 imagery (2.5 m

resolution) and $90,000 for IKONOS/Quickbird (<1 m resolution). [accuracy is

likely to be higher for IKONOS/Quickbird, but it is not clear if that would be

worth the price premium. In reality, due to data availability, it may be

necessary to use a combination of the two, and maybe other sensors too].

1.6.2. Data interpretation and processing is high with hyperspatial data, and with

mosaicking large volumes of Landsat data for colour balancing and cloud

removal. This task could probably be done in Liberia with a team of 3-4 GIS

specialists working for a year, with appropriate training and support, and

access to commercial software licenses such as ENVI (c. $6000/license,

discounts may be available). Alternatively an external consultant could

commit to this for about 200-300 skilled days work.

Recommendations for activity data mapping from 2016 onwards

While it would be possible to rely on the Global Forest Watch (Hansen et al. 2013) product,

used for calculating the Activity Data for Liberia’s Reference Level, for MRV, this has various

disadvantages.

a) The data are released about a year in arrears – for example the site56 that allows a user

to download the raw data (rather than just visualise it as with Global Forest Watch) is

still giving 2014 as the final year of deforestation available, and it is now 8 months after

2015 finished.

b) There is no guarantee that these data will be produced in a similar manner in perpetuity,

so relying on them would be a substantial risk.

c) The 30 m resolution is quite coarse, especially compared to the 10 m resolution

Metria/Geoville map for 2014 – this may reduce the accuracy of the results.

d) The deforestation data is binary, reporting forest loss only regardless of starting class.

No classification by canopy cover has been performed in this dataset since 2000. Liberia

will need to perform a process of stratification, updated frequently, in order to assign

56 http://earthenginepartners.appspot.com/science-2013-global-forest/download_v1.2.html

http://earthenginepartners.appspot.com/science-2013-global-forest/download_v1.2.html



106

losses to different strata: the Hansen et al. data will not provide this stratification, so

further mapping will be necessary.

e) No annual gain data layer is given by Hansen et al. There is a gain product, but it is not

given a year, unlike for deforestation. Thus it is not possible to produce a net

deforestation figure from these data, which would not allow for complete reporting

under IPCC and MGD guidance.

f) Forest loss can only occur in a pixel once, so a pixel that was deforested in 2001 is never

again reported as deforested. As the length of the dataset increases this is becoming

ever more of a problem, especially in an area with rapid tree growth such as Liberia. This

will over time lead to an under-reporting of deforestation rate if these data are used, as

deforestation in pixels that have regrown into forest, and then are deforested again, will

not be reported.

As a consequence we would not recommend continuing to use the Hansen et al. dataset for

monitoring forest loss in Liberia in the future. Here we present two possible alternative methods

that are recommended (Box 5).

Two approaches for monitoring forest cover change over time into the future

Option 1:

1 Optical data at 10 m resolution

Geoville & Metria successfully produced a landcover map for Liberia in 2014 using a

combination of 5 m resolution RapidEye and 30 m resolution Landsat 8 data. The most obvious

way to monitor deforestation, and more general land use change, in Liberia, would therefore

be to continue the methods they have used successfully.

RapidEye is a commercial satellite and charges for data; Landsat 8 is free but its 30 m resolution

is relatively coarse, and no successor satellite has yet been fully funded by the US Government.

Landsat 7 is still operational but it has a significant problem in terms of stripes in the data

caused by its Scan Line Corrector failing in 2003, therefore if Landsat 8 was to fail there would

be a significant data gap. Landsat 9, if funded, is currently expected to launch in 2023, 5 years

after the 5-year design lifetime of Landsat 8. A potential solution to the cost problem with

RapidEye, and resolution/data supply risk of Landsat, has emerged with the launch of Sentinel

2. This is an EU-funded set of operational satellites, part of the Copernicus program, which

commits to having two of each operational satellites in orbit continuously well into the 2030’s,

with continuity of design and data supply. Like Landsat, the data are to be distributed in an

unrestricted format.



107

Sentinel 2 is a 10 m resolution satellite with bands similar to those on Landsat (more extensive

than those on RapidEye), and a frequent revisit time (10 days currently, every 5 days once two

satellites are operational in 2017, compared to every 16 days for Landsat 8). This combination

of high resolution, zero data cost, frequent revisits, and a commitment to data provision over

a long period, makes Sentinel 2 our recommended sensor for monitoring Liberia’s forests.

A recommended method follows below

1.7. Sentinel 2 data collection and pre-processing

1.7.1. The frequent observations of Sentinel 2 allows for data to be collected during

a single season. We recommend downloading all scenes with a cloud cover

<20 % during January and February of each year, starting with 2016. This 2-

month window is chosen partly as it is during the height of the dry season in

Liberia, when the difference between forest and non-forest pixels should be

at their greatest, and partly for reporting ease: reporting change January-

January will allow for the fastest possible reporting of a previous year’s

deforestation.

1.7.2. Scenes should be downloaded at Level 2A. There are various data hubs

available for data download, it may be to Liberia’s advantage to set up an

International Data Hub agreement with the European Union, but it could

otherwise use one of the other data hubs for example the general

Scientific/Other Use Data Hub57

1.7.3. The scenes should be mosaicked, choosing for each pixel the earliest possible

cloud free observation. Clouds and cloud-shadow should be identified using

the standard Quality Indicators which flag clouds and cloud shadows. Simple

software could be developed to automatically download the data and

perform the step of choosing the most recent cloud free pixel, for example

using open source software such as the Sentinel Toolbox and GDAL, held

together using python code.

1.7.4. It may be that, for a small subset of pixels, no cloud free observations were

made in February. In that case targeted download of scenes from March, and

the previous December might be necessary to fill in the gaps.

1.8. Collecting training data

57 https://scihub.copernicus.eu/

https://scihub.copernicus.eu/



108

1.8.1. At its most basic all that is needed for training data is a set of points

collected in the January of each year stating whether they are ‘forest’

or ‘non-forest’. There are options for collecting these data:

a) These could be physical assessments in the field, involving

teams of two people with a GPS, a camera and a means of

assessing canopy cover (for example a curved mirror or

hemispherical lens for the camera). Ideally they would

collect points randomly across the country, but a series of

say 200 points within a set of 10 ‘super-sites’ (10km x 10 km

areas containing a range of forest types) spread across the

country would be sufficient

b) Alternatively these could be marked on an image through

the use of <1.5 m resolution satellite data, where individual

tree canopies are visible. This might be cheaper than option

a) and could be as, if not more, accurate. As with a),

potentially 10 x 10km x 10km areas could be spread across

the country, with data collected from the Pleides or

Worldview constellations

1.8.2. Alternatively, more ambitiously, this task could involve a full landcover

mapping exercise. In this case a set of about 8 classes, including 2-3

forest classes, could be attempted, duplicating more closely the

Geoville/Metria classification with the benefit of being able to track the

land use post deforestation. This would only be possible with a large

scale field campaign, involving many more teams and points than for

1.2.1. It is not likely that a remote-sensing only campaign would be

possible here (as in b. above), because land use is hard to assess

without ground knowledge.

1.9. Classification

1.9.1. Classification would be performed using the annual Sentinel 2 mosaic

and the ground truth data. Further ‘truth’ data points could be added

to the process from obvious classes, for example from water bodies

(lakes, rivers), and obvious non-vegetated areas.



109

1.9.2. While the classification should be performed using the 10 m resolution

data, we would recommend a Minimum Mapping Unit (minimum

possible object size) of 1 ha, to match with Liberia’s forest definition.

Experimentation may show that this should be reduced to say half a

hectare for non-forest classes for best results.

1.9.3. The best classification result would probably come from the use of

proprietary software, for example the use of the Support Vector

Machine or Neural Network algorithms in ENVI58, or an object-based

approach using Ecognition59. However very good results could probably

be achieved from using open source software such as R (where

RandomForest60 or other tools could be used to perform a robust

machine learning classification), or using the Semi-Automated

Classification plugin61 in QGIS.

1.9.4. The map accuracy should be tested by withholding 50 % of the training

data for testing purposes. A forest/non-forest accuracy of >95 % should

easily be achievable, ideally >99%. If a wider landcover classification is

attempted (e.g Option 1.2.2.), the forest/non-forest classification

should still exceed 95%, with accuracy for classes below that >90%.

1.10. Validation

1.10.1. As stated above at the same time as the test dataset is collected,

enough points would be collected for an independent test dataset for

the classification. Typical best practice is to hold back 50 % of points for

training and 50 % for testing. This will provide an accuracy assessment

for the classification. However, it does not validate the change map

performed by differencing the classifications produced in 2016 to 2015,

and so on, as most of the ground truth points will not have changed

class in that period.


targeted field visits to a random set of pixels highlighted by the dataset

58 http://www.exelisvis.co.uk/ProductsServices/ENVIProducts/ENVI.aspx 59 http://www.ecognition.com/ 60 https://cran.r-project.org/web/packages/randomForest/index.html 61 https://plugins.qgis.org/plugins/SemiAutomaticClassificationPlugin/

http://www.exelisvis.co.uk/ProductsServices/ENVIProducts/ENVI.aspx

http://www.ecognition.com/

https://cran.r-project.org/web/packages/randomForest/index.html

https://plugins.qgis.org/plugins/SemiAutomaticClassificationPlugin/



110

as ‘changed’, and a nearby random set of pixels that are ‘unchanged’.

The number of changed and unchanged pixels visited should be equal,

and we would recommend at least 500 points of each type are visited,

spread across ideally 10 areas of the country.

1.10.3. In each case the visited point would be located using a GPS, and

photographs and a canopy cover assessment made to decide if the

point was forest or non-forest. Signs of obvious recent human activity,

for example tree stumps or burn scars, would be recorded.

1.10.4. This will allow a calculation of the error and bias of the deforestation

product


1.11.1. Provided sufficient investment in the physical and network

infrastructure, and training, the above protocol could be performed

entirely within Liberia. This would have the biggest long-term benefits

by developing data processing expertise and ensuring data ownership

within Liberia. However, the costs for this option are likely to be high.

1.11.2. Alternatively, an external consultant could provide the mapping

services following this protocol, and oversee and provide training for

the fieldwork component.

1.11.3. Either way, the initial set-up costs in terms of setting up a software

pathway and physical infrastructure could be high and hard to

estimate. Once set up though, the fieldwork component could be

conducted using existing staff in the FDA or other Liberians at relatively

low cost (perhaps 150-200 person days, plus transport and subsistence

costs), and the data processing task would involve perhaps 120 days of

two GIS Technicians plus 20 days of Management.

Option 2:

2. L-band radar data at 20 m resolution

L-band radar from the ALOS PALSAR (2007-2011) and ALOS-2 PALSAR-2 (2014-) satellites has

been successfully used for mapping aboveground biomass, degradation and deforestation in



111

landscapes similar to those found in Liberia (e.g. in neighbouring Sierra Leone62, in forest-

agriculture mosaics in Peru63, ). However, this more sophisticated use does not prevent it being

used solely to reliably map simple change from forest to non-forest classes. Indeed, JAXA

produced a well-regarded forest/non-forest product from ALOS PALSAR for 2007-2011 at an

impressive 10m resolution64: it is not well calibrated for Liberia’s forests, but suggests that such

a product could be easily produced.

ALOS-2 PALSAR-2 data is not freely available, and thus data would have to be purchased.

However, it does have various advantages over optical data: it does not suffer from cloud

cover as the sensor can see through clouds, and it is only sensitive to trees, with grasses and

crops having little effect. Therefore, if well calibrated, it can produce very high quality maps

of forest and non-forest suitable for detecting deforestation at high accuracy.

A recommended method follows below

2.1. ALOS-2 PALSAR-2 Data download and pre-processing

2.1.1. We recommend that the first scenes collected across Liberia each year are

ordered and downloaded. It is expected that a complete mosaic will be

possible using scenes collected in January and early February.

2.1.2. Scenes should be ordered in the ‘Fine’, 9.1 x 5.3 m resolution mode, as 70 x

70 km tiles. Level 1.1. data should be ordered for best results, though if

capacity (physical and digital) is not high enough level 2.1 data, which is

terrain corrected and geocoded already, would reduce download size and

processing without greatly reducing accuracy.

2.1.3. At the time of writing, the cost per scene is YEN300,000 from http://en.alos-

pasco.com/offer/price.html, approximately $3,000. However we would

expect a discount of at least 30 % would be available given the high volume

and use of the data, and it is possible data would be provided free of charge

through Japanese aid programs or their Kyoto and Carbon program, given the

use case here.

62 http://database.v-c-s.org/sites/v-c-s.org/files/VCS+CCB+MonitoringImplementationReport%20SUBMITTED%2025Aug2015.pdf 63 http://iopscience.iop.org/article/10.1088/1748-9326/10/3/034014/meta 64 http://www.eorc.jaxa.jp/ALOS/en/palsar_fnf/fnf_index.htm http://www.sciencedirect.com/science/article/pii/S0034425714001527

http://en.alos-pasco.com/offer/price.html

http://en.alos-pasco.com/offer/price.html

http://database.v-c-s.org/sites/v-c-s.org/files/VCS+CCB+MonitoringImplementationReport%20SUBMITTED%2025Aug2015.pdf

http://database.v-c-s.org/sites/v-c-s.org/files/VCS+CCB+MonitoringImplementationReport%20SUBMITTED%2025Aug2015.pdf

http://iopscience.iop.org/article/10.1088/1748-9326/10/3/034014/meta


http://www.sciencedirect.com/science/article/pii/S0034425714001527



112

2.1.4. 34 scenes are required to cover Liberia, and all should be downloaded for

January 2015, January 2016, etc.

2.1.5. Data pre-processing would involve terrain correcting, multilooking and

geocoding the data using open source software such as the Sentinel-1-

toolbox65, or commercial software such as SARscape within ENVI66 or

GAMMA67. The 30 m SRTM DEM should be used, unless another is available.

As stated in 2.1.2, if the capacity is not available to use these software, then

data download at level 2.1 would skip this step, and the processed scenes

could be directly mosaicked using software such as QGIS or ArcMap. However

the terrain correction and data quality would be lower.

2.2. Collecting training data

2.2.1. At its most basic all that is needed for training data is a set of points

collected in the January of each year stating whether they are ‘forest’

or ‘non-forest’. There are options for collecting these data:

c) These could be physical assessments in the field, involving

teams of two people with a GPS, a camera and a means of

assessing canopy cover (for example a curved mirror or

hemispherical lens for the camera). Ideally they would

collect points randomly across the country, but a series of

say 200 points within a set of 10 ‘super-sites’ (10km x 10 km

areas containing a range of forest types) spread across the

country would be sufficient

d) Alternatively these could be marked on an image through

the use of <1.5 m resolution satellite data, where individual

tree canopies are visible. This might be cheaper than option

a) and could be as, if not more, accurate. As with a),

potentially 10 x 10km x 10km areas could be spread across

the country, with data collected from the Pleides or

Worldview constellations

65 http://step.esa.int/main/toolboxes/sentinel-1-toolbox/ 66 http://www.sarmap.ch/page.php?page=sarscape 67 http://www.gamma-rs.ch/no_cache/software.html

http://step.esa.int/main/toolboxes/sentinel-1-toolbox/

http://www.sarmap.ch/page.php?page=sarscape

http://www.gamma-rs.ch/no_cache/software.html



113

2.3. Classification

2.3.1. Classification would be performed through a simple threshold on the

HV polarisation data: the ground truth data would be used to assign a

threshold HV value where a pixel is forest, and all pixels below that

point would be classified as ‘non-forest’ and those above would be

‘forest’.

2.3.2. Half the ground data should be used to develop this threshold, the

other half to test it. If the accuracy is not sufficiently high (<95%), then

the landscape should be split using the stratification, performed

elsewhere, and the classification process performed separately for the

different strata. We would then expect an accuracy in excess of 95%

compared to the ground data.

2.3.3. The classification should be performed at 20 m resolution, but post-

processing after the classification should remove any remaining ‘forest’

patches with a size smaller than 1 ha, as these are not, in fact, forest,

by Liberia’s definitions.

2.3.4. A comparison of the annual maps will give areas of deforestation and

regrowth.

2.4. Validation

2.4.1. As stated above at the same time as the test dataset is collected,

enough points would be collected for an independent test dataset for

the classification. Typical best practice is to hold back 50 % of points for

training and 50 % for testing. This will provide an accuracy assessment

for the classification. However, it does not validate the change map

performed by differencing the classifications produced in 2016 to 2015,

and so on, as most of the ground truth points will not have changed

class in that period.


targeted field visits to a random set of pixels highlighted by the dataset

as ‘changed’, and a nearby random set of pixels that are ‘unchanged’.

The number of changed and unchanged pixels visited should be equal,

and we would recommend at least 500 points of each type are visited,

spread across ideally 10 areas of the country.



114

2.4.3. In each case the visited point would be located using a GPS, and

photographs and a canopy cover assessment made to decide if the

point was forest or non-forest. Signs of obvious recent human activity,

for example tree stumps or burn scars, would be recorded.

2.4.4. This will allow a calculation of the error and bias of the deforestation

product


2.5.1. Assuming a 30% volume discount, the data cost would be $68,000 per

year. It might be possible to reduce this cost significantly, and possibly

completely, through negotiations with JAXA.

2.5.2. Licences for commercial software such as GAMMA or SARscape are

likely to be on the order of $10K, though again discounts could be

available. Free software could be used, though results are likely to be

less good.

2.5.3. Implementing the processing, if software and trained staff existed,

would not be difficult, either internally or through an external

consultant. It would perhaps be 60 days’ work for a trained technician,

involving 30-50% as much time as the optical method in 1.

2.5.4. Fieldwork costs are hard to estimate, but would be identical to method

1. Future fieldwork costs for calibration are likely to be lower than

Method 1, as there are not expected to be calibration differences from

year to year.



115

APPENDIX F: MAPPING LAND USE WITH REMOTE SENSING

Satellites detect properties of the vegetation, soil, topography and buildings on a piece of land. These may

relate to land use, but the features measured by remote sensing are rarely unique to a particular landuse

type. As a typical example, a grassland area grazed by cattle, and one grazed only by wild animals, could

look identical to a satellite. Therefore land use mapping typically relies on a combination of satellite data

(which mainly gives landcover) and ancillary data, such as vector layers giving the boundaries of

plantations, or the distance from roads. Often the eventual landuse map is as much a model as a set of

observations.

One other technique that is often used is the use of historical data to assist with current land use mapping.

For example an area of bright green trees might be either a rubber plantation or a forest; but if it was

known to be non-forest 5 years ago, it might become certain that it was a forest. Similarly, a small patch

might look like secondary forest, but if it has been cleared 3 times in the past 15 years, then it is likely to

be under swidden agriculture cycles.

Finally, improved maps of land use can be developed through combining different types of remote sensing

data. For example, optical data gives information on vegetation greenness and could provide historical

data, and long-wavelength radar could provide a biomass map: a fusion of the two could produce more

accurate land use mapping. Fusion further with vector data might provide a highly accurate map.

Proposed methodology

A proposed methodology for land use mapping in Liberia could follow these steps:

1. Collate input raster layers

1.1. Optical layers of landcover (ideally those produced from a different analysis, for example that

connected with Liberia’s Activity Data). These should cover at least a decade, ideally with four

points, e.g. 2003, 2007, 2010, 2013, 2016 would be ideal. The maps produced during the

Activity Data task using the Metria/Geoville map and the Hansen et al. deforestation data

would provide a reasonable starting point.

1.2. Collate currently optical satellite data. This could for example be a mosaic of Landsat 8 data, or

Sentinel-2, from the same season.

1.3. [Collate current radar satellite data, for example by creating a mosaic of PALSAR-2 data (see AD

proposed methods) – this is optional, and could maybe be added if initial accuracy not

sufficiently good]

1.4. All layers should be warped to the same projection and to match each other, using the highest

available resolution and most recent data as the reference image (e.g. 10 m Sentinel-2).



116

2. Collate input vector layers

2.1. Collate all vector layers that could potentially be useful, including at least

2.1.1. Settlements and their populations

2.1.2. Roads (including quality)

2.1.3. Rivers

2.1.4. The concession boundaries for all land use types that are available

2.2. Reproject to the same projection/datum as the optical data (typically UTM), and compare to

high resolution optical data to ensure e.g. roads and rivers line up with the data. If not, adjust.

2.3. Create raster layers from the road and settlement layers, giving layers such as ‘distance from

road’, ‘travel time to the nearest village’, ‘travel time to the nearest town’, ‘travel time to the

nearest city’, ‘travel time to the nearest sea port’ or similar. Consult local experts as to what

would be the most relevant cut-off points for community size here for different economic

activities (e.g. selling charcoal vs timber), and to mean travel time on different qualities of roads

(and rivers if appropriate.)

3. Collect ground data on actual land use

3.1. At least 200 examples of each land use type should be visited and recorded with a GPS (outline

of areas of at least 0.5 ha). Ideally these should be spread through the country.

3.2. Again expert guidance may be needed as to what constitutes the most important land use type

and how they can be found.

4. Classification and validation

4.1. Half the input points should be used to train a neural network algorithm, involving all input data

layers (including radar if available, historical classifications, vector-derived rasters, as well as

current optical data). This could be performed using QGIS, but commercial software such as

ENVI or object-based software such as Ecognition would probably produce the best results.

4.2. The map should be validated against 50 % of data held back for testing.



117

ANNEX 1: FOREST DEGRADATION

Degradation Emission Estimates

Forest degradation involves greenhouse gas emissions taking place within forest areas as a result of

anthropogenic actions, but with the area remaining with the national forest definition throughout.

Degradation can include things such as commercial selective logging for timber, small-scale legal and illegal

logging for village use and timber, fuelwood collection, and fire. Drivers of forest degradation impact

forests in a variety of ways, and result in different magnitudes of emissions. The primary activities leading

to deforestation and forest degradation in Liberia, as identified by Liberia’s R-PP and discussed in the Land

Use Analysis (LTS 2016), are forestry, agriculture, mining, and charcoal production, all at both the

commercial and community scale. The MRV Roadmap distinguishes between the drivers of deforestation

and degradation, and lists commercial logging, chainsaw logging, conversion of natural forests to forest

plantations, fuel wood collection and charcoal production as drivers of degradation.

There are two fundamentally different general approaches to measuring and monitoring emissions from

forest degradation, both of which are briefly discussed below.

The activity-based approach focuses on specific forest degradation activities, such as timber harvesting

or woodfuel collection and allows accounting to focus on the forest degradation activities assumed to

have the largest impact on degradation emissions. Under this approach, specific emission factors and

activity data are generated for each forest degradation activity included in the REDD+ program.

The land-based approach can be taken using remote sensing products to detect where forest cover

decreases, and therefore is assumed to be forest degradation. Under this approach, the specific source

or driver of forest degradation is not a particular consideration, and rather emissions are estimated based

on the difference between carbon stocks in the before forest degradation and after forest degradation

scenarios.

Regardless of the method used, identifying forest degradation is a considerably more difficult technical

challenge than identifying deforestation. Degradation estimates can be developed using a variety of

spatial and non-spatial datasets.

Currently, there are no reliable country-specific data on the extent of degradation that can be used to

estimate historical emissions from degradation for use in the REL. The descriptions below merely provide



118

an indication of the potential magnitude of degradation emissions, and the general methods that were

used to derive these estimates.

Approximating degradation magnitude by ‘activity’

Estimating emissions from forest degradation based on activities requires knowledge of the prevalent

degrading activities, and data on both the extent of these activities and the resulting emissions.

Unfortunately, there is currently no available data suitable for assessing rates or the extent of degradation

or that can be used to estimate either the emissions per unit of degrading activity.

First order estimates of Liberia’s emissions have been assessed by activity using the methods developed

by Winrock for the World Bank REDD+ Decision Support Toolbox68 (Table A1-1). These estimates reflect

Liberia’s forest definition69 and can largely be considered IPCC Tier-2 due to their use of global spatial

datasets and country-specific data. A full description of the methods and data sources to produce

estimates of emissions per activity are included in below.

Table A1-1. First order estimate of Liberia's emissions for deforestation and forest degradation using data from WB Decision

Support Tool. Forest degradation is comprised here of logging, woodfuel and fire. (Note that deforestation estimates are based

on the DST, not the methods described elsewhere in this report, and are only shown here for comparison to degradation.)

Activity tCO2 emissions per

year

Percent of total

emissions

Deforestation 18,946,559 91.3%

Degradation by activity

Logging 1,320,835 6.4%

Woodfuel 217,835 1.0%

Fire 272,436 1.3%

Total Emissions 20,757,665 100%

Enhancements -378,136 --

These estimates indicate that deforestation is the source of the vast majority of Liberia’s emissions from

the forest sector. While this is likely true, the estimates provided in Table A1-1 are based on global

datasets, which have limitations.

68 Sidman, G., L. Murray, T.R.H. Pearson, N.L. Harris, M. Netzer. 2014. World Bank REDD+ Decision Support Toolbox Methods. Online DST available at http://redd-dst.ags.io. 69 Canopy closure exceeding 30%

http://redd-dst.ags.io/



119

If an activity based approach is utilized to estimate Liberia’s emissions from forest degradation, relevant

activity data includes volume and type of timber harvested, mill efficiency, and amount of woodfuel

harvested for heating or cooking. These data may be impossible to acquire if they do not already exist,

and while they may be more accurate than remotely sensed activity data, they are contingent upon

complete accounting and record-keeping70.

Methods and data sources applied for generating first order estimates of deforestation and degradation

emissions

A set of methods were developed to create estimates of emissions from deforestation and the main

degradation activities. The methods follow the approach developed for the World Bank funded REDD+

Decision Support Toolbox (REDD+ DST)71, available online. The information below has been adapted from

the full description of methods applied in this Toolbox72.

Deforestation

Activity Data

Hansen et al. (2013)73 raster layers, which were derived from Landsat 7 ETM+ satellite images, were used

for all activity data for deforestation in the REDD+ DST. The tree cover raster, which shows “canopy closure

for all vegetation taller than 5 m in height” as a percentage from 0-100, was used to establish area of

forest. The REDD+ DST allows users to select either 10%, 20%, or 30% canopy cover as the definition of

forest, but estimates offered in this report reflect the 30% canopy cover threshold, to match with Liberia’s

national forest definition. Cells with values of equal to or greater than each canopy cover threshold were

extracted from the original tree cover raster to create forest mask for each canopy cover definition.

The Hansen et al. (2013) loss year raster was then used to determine areas of deforestation. The loss year

raster shows all areas that were deforested, on an annual basis, between 2001-2012. Areas of

deforestation for each year between 2001-2012 were clipped to each forest definition threshold (10%,

70 Goslee, K.M., S. Walker, S. Brown, T.R.H. Pearson, P. Stephen, R. Turner, and A.M. Grais. 2015. Technical Guidance Series for the Development of a National or Subnational Forest Monitoring System for REDD+: Forest Degradation Guidance and Decision Support Tool. Developed by Winrock International and the United States Forest Service under the USAID LEAF Program 71 http://www.forestcarbonpartnership.org/dst 72 https://redd-dst.ags.io/static/docs/REDD%2B%20DST%20Methods%20and%20Data%20Sources.pdf 73 Hansen, M. C., P. V. Potapov, R. Moore, M. Hancher, S. A. Turubanova, A. Tyukavina, D. Thau, S. V. Stehman, S. J. Goetz, T. R. Loveland, A. Kommareddy, A. Egorov, L. Chini, C. O. Justice, and J. R. G. Townshend. 2013. “High-Resolution Global Maps of 21st-Century Forest Cover Change.” Science 342 (15 November): 850–53. Data available on-line from: http://earthenginepartners.appspot.com/science-2013-global-forest.

http://www.forestcarbonpartnership.org/dst

https://redd-dst.ags.io/static/docs/REDD%2B%20DST%20Methods%20and%20Data%20Sources.pdf




120

20%, and 30%) canopy cover) and summed for each subnational administrative unit.

Subnational units used in the WB DST were derived from the Database of Global Administrative Areas

(GADM)74 which provides a data layer of national and subnational political boundaries. Both Level 1 and

Level 2 subnational units are available in the WB DST, which represent subnational boundaries at different

levels of scale. GADM Level 1 units are typically states, departments, or prefectures whereas Level 2 GADM

units subdivide Level 1 units into municipalities or counties. The first order estimates of emissions from

deforestation produced by the REDD+ DST are the average deforestation emissions for each year Hansen

et al. (2013) activity data are available (2001-2012).

Carbon Stocks/Emission Factors

Estimates of emissions from deforestation in the DST are inclusive of all relevant carbon pools:

aboveground biomass, belowground biomass, deadwood and litter, and soil carbon. All emissions from

deforestation except for those from the soil/peat pool are assumed to be committed the year that the

deforestation activity occurs. The contribution of emissions from the soil pool are calculated differently as

post-deforestation land use and soil type must also be considered, according to Intergovernmental Panel

on Climate Change (IPCC) Guidelines. Biomass values were converted to carbon values by dividing biomass

in half, and carbon dioxide values were then estimated by applying the molecular weight ratio of carbon

dioxide to carbon (i.e., carbon estimates were multiplied by 44/12).

Aboveground Biomass values were obtained from a spatial layer of carbon stocks in tropical areas

developed by Saatchi et al. (2011)75 which maps aboveground biomass carbon stocks per hectare over

Latin America, Africa, and Asia for the early 2000s, providing a useful pre-deforestation benchmark for the

REDD+ DST. The biomass map was clipped to three forest mask layers matching the three forest definitions

used in the DST (10%, 20%, and 30% canopy cover). Since the forest canopy layer (Hansen et al. 2013) and

the biomass layer (Saatchi et al. 2011) use different remote sensing sources and have different spatial

resolutions, there were some inevitable mismatches between the two data sources. To prevent counting

non-forest biomass pixels that were retained after clipping to the forest masks, all pixels that had less than

40 tons of aboveground carbon per hectare were removed. This biomass threshold has been used to

exclude pixels represented as forest in the Hansen dataset but effectively not considered as such in the

Saatchi dataset from biasing carbon stocks for a broader deforested area (40 t C/ha derived from

74 Available on-line from: www.gadm.org 75 Saatchi, S.S., Harris, N.L., Brown, S., Lefsky, M., Mitchard, E.T.A., Salas, W., Zutta, B.R., Buermann, W., Lewis, S.L., Hagen, S., Petrova, S., White, L., Silman, M., Morel, A. 2011. “Benchmark map of forest carbon stocks in tropical regions across three continents.” Proceedings of the National Academy of Sciences, USA, 108, 9899.



121

Dinerstein et al. 2014)76. Although this threshold may not be accurate in other forest biomes that exist in

countries shown in the REDD+ DST, it provided a conservative estimate of aboveground biomass that could

be applied universally across forested areas. The resulting biomass was averaged across subnational units

using a zonal statistics function, and then converted from biomass to tons of carbon by dividing in half, as

specified in Intergovernmental Panel on Climate Change (IPCC) Good Practice Guidance for Land Use,

Land-Use Change and Forestry (LULUCF)77.

Belowground Biomass estimates were developed through the application of an allometric equation

developed by Mokany et al. (2006)78:

𝐵𝐺𝐵 = 0.489𝐴𝐺𝐵0.89

Deadwood and Litter estimates were calculated based on a fraction of aboveground biomass as specified

by methods under the United Nations Framework on Climate Change’s (UNFCCC)

Afforestation/Reforestation Clean Development Mechanism (A/R CDM)79. This methodology assumes

deadwood and litter to be a fraction of aboveground biomass based on an area’s elevation and annual

precipitation regime (Table A1-1). Only fractions for tropical biomes were used in the REDD+ DST.

Elevation was obtained from the Global 30 Arc-Second Elevation (GTOPO30)80 digital elevation model and

the annual precipitation from the WorldClim database81.

76 Dinerstein, E., Baccini, A., Anderson, M., Fiske, G., Wikramanayake E., McLauglin, D., Powell, G., Olson, D., Joshi, A. 2014. “Guiding Agricultural Expansion to Spare Tropical Forests.” Conservation Letters, in press. 77 IPCC. 2003. Good Practice Guidance for Land Use, Land-Use Change and Forestry. Penman, J., Gytarsky, M., Hiraishi, T., Krug, T., Druger, D., Pipatti, R., Buendia, L., Miwa, K., Ngara, T., Tanabe, K., Wagner, F. (eds). Published: IGES, Japan. Available online at http://www.ipcc-nggip.iges.or.jp/public/gpglulucf/gpglulucf.html. 78 Mokany, K., Raison, J.R., Prokushkin, A.S. 2006. Critical analysis of root : shoot ratios in terrestrial biomes. Global Change Biology, 12, 84-84, doi: 10.1111/j.1365-2486.2005.001043.x. 79 UNFCCC. 2012. “Estimation of carbon stocks and change in carbon stocks in dead wood and litter in A/R CDM project activities Version 2.0.0.” EB 67 Report Annex 23. 80 United States Geological Survey. “Global 30 Arc-Second Elevation (GTOPO30). Available online at https://lta.cr.usgs.gov/GTOPO30. 81 Hijmans, R.J., S.E. Cameron, J.L. Parra, P.G. Jones and A. Jarvis, 2005. “Very high resolution interpolated climate surfaces for global land areas.” International Journal of Climatology 25, 1965-1978.



122

Table A1-1: UNFCCC A/R CDM methodology for determining deadwood and litter biomass stocks from aboveground biomass.

Numbers shown are for the tropical biome only.

Soil Carbon82 emissions were estimated leveraging data from the Harmonized World Soil Database which

offers data on carbon content (in the top 30 cm of soil, which is the assumed depth affected by

deforestation) and bulk density. To estimate tons of carbon per hectare in forested areas, the bulk density

was multiplied by the volume of topsoil in one hectare, and then multiplied by the fraction of carbon

content. This was done for all pixels within the forest mask, and the weighted average was found for each

subnational unit.

To calculate soil emissions from deforestation, land use change factors (FLU) were used. FLUs were obtained

from the IPCC Guidelines for National Greenhouse Gas Inventories83. Only FLUs for conversion to long-term

cultivated crops were used, which varied based on the temperature regime of the subnational unit

(tropical or temperate). Although not all land will become long-term cultivated crops, this assumption was

made in the absence of a good method of predicting post-deforestation land use on the local level across

all FCPF countries. The following formula was used to find deforestation emissions EmsSOIL from soil carbon

based on pre-deforestation soil carbon stocks (CPRE):

𝐸𝑚𝑠𝑆𝑂𝐼𝐿 = 𝐶𝑃𝑅𝐸 − (𝐶𝑃𝑅𝐸 ∗ 𝐹𝐿𝑈)

82 The REDD+ DST estimated soil carbon emissions from deforestation on mineral soils differently from those on peat, but as no emissions from peat soils were included in estimates for Liberia, methods are not described here. 83 IPCC. 2006. “2006 IPCC Guidelines for National Greenhouse Gas Inventories.” Prepared by the National Greenhouse Gas Inventories Programme, Eggleston H.S., Buendia L., Miwa K., Ngara T. and Tanabe K. (eds). Published: IGES, Japan. Volume 4 Agriculture, Forestry and Other Land Use. Paustian, K, Ravindranath, N.H. and Van Amstel, A (coordinating lead authors). Available at: http://www.ipcc-nggip.iges.or.jp/public/2006gl/vol1.html.

1. ELEVATION

(m)

2. ANNUAL

PRECIPITATION

(mm yr-1)

3. DEADWOOD

FRACTION OF

AGB

4. LITTER

FRACTION

OF AGB

5. < 2000 6. < 1000 7. 0.02 8. 0.04

9. < 2000 10. 1000 – 1600 11. 0.01 12. 0.01

13. < 2000 14. > 1600 15. 0.06 16. 0.01

17. > 2000 18. All 19. 0.07 20. 0.01



123

Forest Degradation

Due to the increased complexity associated with detecting and measuring the impacts of activities that do

not result in deforestation, but degrade forest carbon stocks, methods and data sources applied in the

REDD+ DST to calculate deforestation emissions differ considerably from those applied to calculate

emissions resulting from degradation activities.

Timber Harvesting

Activity Data: In the context of the REDD+ DST, timber harvesting refers to commercial selective logging.

The methodology described in Pearson et al. (2014) 84 was used to calculate national-level logging

emissions. This methodology used extraction volumes from the 2010 FAO Global Forest Resources

Assessment, and then calculates emissions from extracted logs, damage to the surrounding trees at the

logging location, and logging infrastructure.

As emissions were calculated on a national scale, it was necessary to divide emissions among the

subnational units represented in the REDD+ DST. The Global Forest Watch database85 provides logging

concessions data for Liberia, and thus national-level logging emissions were divided according to the

proportion of national concessions area within Liberia’s subnational units.

Emission Factors: Some of the volume in extracted logs is stored as harvested wood products (HWP) in

the form of lumber, wood panels, or other products that have an in-use lifetime and then may remain

sequestered even after disposal especially when in landfills. As such, harvested wood that is manufactured

into these products does not immediately contribute to emissions, so the portion of extracted wood that

is effectively permanently sequestered in HWP must be subtracted from total logging emissions. Storage

at 100 years is used as a simplification for permanent storage reflecting estimations of atmospheric

residence of carbon dioxide. Earles et al. (2012)86 calculated the percentage of aboveground carbon in

harvested timber that remains stored in HWP after 100 years by estimating the proportion of national-

level extraction volume data that became long lasting end products.

To calculate the amount of aboveground carbon stored in HWP for each subnational unit (AGBHWP), the

following equation was used where LNAT is national-level logging extraction volume, PctHWP is the percent

84Pearson, T.R.H., Brown, S., Casarim, F.M. 2014. “Carbon emissions from tropical forest degradation caused by logging.” Environmental Research Letters, 9, 034017. doi:10.1088/1748-9326/9/3/034017 85“Logging.” World Resources Institute. Accessed through Global Forest Watch on Oct 7 2014. Available online at www.globalforestwatch.org. 86Earles, J.M., Yeh, S., Skog, K.E. 2012. “Timing of carbon emissions from global forest clearance.” Nature Climate Change, 2, 682-685. doi:10.1038/nclimate1535



124

of aboveground biomass stored in HWP after 100 years, BCEF is the biomass conversion and expansion

factor, and FArea is the proportion of national forested area within the subnational unit:

𝐴𝐺𝐵𝐻𝑊𝑃 = 𝐿𝑁𝐴𝑇 ∗ 𝑃𝑐𝑡𝐻𝑊𝑃 ∗ 𝐵𝐶𝐸𝐹 ∗ 𝐹𝐴𝑟𝑒𝑎

Biomass conversion and expansion factors (BCEF) allow for the conversion of merchantable growing stock

volume to aboveground biomass. BCEFs for temperate conifer forests and humid tropical natural forests

were used from the 2006 IPCC Guidelines for National Greenhouse Gas Inventories. For the seven

countries that have concessions data available through Global Forest Watch, the proportion of national

concessions area within the subnational unit was substituted for the proportion of national forested area

within the subnational unit in the above equation.

Woodfuel

To estimate emissions from woodfuel harvesting, an analysis conducted by Drigo et al. (2014)87 was

leveraged to derive estimates that reflect emissions from the fraction of non-renewable woodfuel harvest.

Drigo et al. (2014)’s analysis offered estimates of NRB from land cover change (LCC) by-products was as

well, since some wood that is burned as woodfuel comes from deforestation rather than degradation. In

an effort to avoid double-counting emissions, in the REDD+ DST, only the woodfuel demand that was

satisfied by non-LCC by-products was considered.

Forest Fire

Emissions from forest fires are the third source of degradation emissions included in the REDD+ DST.

Forest fires refer to fires that degrade the forest through low to high severity burning, but are not the

source of fires that cause a land cover change, such as human-induced deforestation. The Global Fire

Emissions Database (GFED)88 provides monthly dry matter emissions that are classified into different

sources and land cover types. Within the humid tropical forest biome, deforestation fire emissions are

decoupled from other emissions based on fire persistence (the length of time for which a fire burns in the

87Drigo, R. 2014. “Elaboration of the pan-tropical analysis of NRB harvesting (Tier 1 data, version 01 April 2014).” Produced by the Yale-UNAM GACC Project: Geospatial Analysis and Modeling of Non-Renewable Biomass: WISDOM and Beyond for Global Alliance for Clean Cookstoves. 88van der Werf, G.R., Randerson, J.T., Giglio, L., Collatz, G.J., Mu, M., Kasibhatla, P.S., Morton, D.C., DeFries, R.S., Jin, Y., van Leewen, T.T. 2010. “Global fire emissions and the contribution of deforestation, savannah, forest, agriculture, and peat fires (1997-2009).” Atmospheric Chemistry and Physics, 10, 11707-11735. doi:10.5194/acp-10-11707-2010



125

same location). Deforestation fires are assumed to have a longer fire persistence in order to achieve

complete combustion of fuels, clearing the land completely for a different land cover use.

To only count emissions from forest fires that contribute to emissions from degradation (since

deforestation fires are already included in the deforestation emissions), only emissions from the forest

land cover class were tabulated in the REDD+ DST. Furthermore, emissions from three main gases were

included: carbon dioxide, methane, and nitrous oxide. Methane and nitrous oxide were converted to

carbon dioxide equivalent, and total emissions per hectare were averaged over each subnational unit.

Carbon stock Enhancements through Afforestation and Reforestation

Afforestation is the establishment of forest on non-forest land that had not previously been forest for a

long period of time while reforestation is the establishment of forest on recently deforested land. Zomer

et al. (2008)89 created a global layer of land suitable for A/R based on several biophysical suitability

variables. These variables excluded lands with high aridity, elevation above tree line, urban areas, water

bodies, areas of high agricultural production, and current/recently deforested areas. Forests were defined

as above 30% canopy cover according to a Moderate Resolution Imaging Spectroradiometer (MODIS)

Vegetation Continuous Fields (VCF) layer of canopy cover. The MODIS VCF canopy cover used in Zomer et

al. (2008) was different than the Hansen et al. (2013) layer used in the REDD+ DST for canopy cover,

resulting in a mismatch of forest definitions. Due to this mismatch, the Hansen et al. (2013) forest masks

for each forest definition were used to clip the Zomer et al. (2008) A/R layer so that only A/R land on non-

forest land was included in enhancements calculations in the REDD+ DST.

Default values for annual average aboveground biomass increments in plantations from the IPCC Good

Practice Guidance for LULUCF were used for annual increases in tons of carbon. Subnational units were

given a biomass increment based on their precipitation regime (from the WorldClim dataset) and location

(Africa, Asia, or the Americas). Biomass increments for tree categories (pine, eucalyptus, etc.) for each

location were averaged to create one biomass increment per continent. The tabulated area of eligible land

for A/R was multiplied by the biomass increment and converted into emissions in tons of carbon dioxide

equivalent, giving each subnational area an annual rate of emissions removals. Since it is not feasible to

convert all land eligible for A/R into forest, the values reported for enhancements in the DST assume a

conversion of 20% of eligible A/R land when considering the removals potential of an area.

89Zomer, R.J., Trabucco, A., Bossio, D.A., Verchot, L.V. 2008. “Climate Change Mitigation: A spatial analysis of global land suitability for clean development mechanism afforestation and reforestation.” International Agricultural Research and Climate Change, 126, 1-2, 67-80.



126

Evaluating potential historical degradation through the use of spatially derived land cover data

As stated, currently the data required to estimate degradation using remote sensing is insufficient for

Liberia. Identifying degradation is challenging, and requires generally requires high resolution spatial data

along a time series together with significant amount of field data (Box 3).

Challenges to identifying degradation using Remote Sensing

Identifying degradation using remote sensing products is more complex than estimating

deforestation, significantly less accurate, and may not thoroughly detect degradation. This is true for

three different reasons:

1. Degradation is more difficult to accurately detect using remote sensing, as by definition a

significant proportion of canopy cover remains after the loss event. If trees are removed

below the canopy, degradation may be entirely invisible to normal, optical remote sensing,

that only sees the top of the canopy. Even if canopy trees are removed, there may only be a

short window (perhaps a few months) where it is visible to optical remote sensing before

regrowth in the gap masks the change.

In general the smaller the magnitude of change (e.g. the fewer canopy trees removed), the

harder it is to detect: i.e. a change from 100% to 35% canopy cover will be more likely to be

detected than a change from 80% to 65%, but both are degradation. Also the size of the patch

of degradation is critical: the larger it is the more likely it will be detected.

For most degradation levels, the pixel resolution and temporal frequency of medium

resolution sensors such as Landsat are insufficient to capture degradation completely90. The

detection accuracy of higher resolution 5-meter imagery is still only around 80%, which points

to the potential limitations of detection using only existing space-borne optical sensors

(Manley et al. 2013). Area data can also be estimated based on ground surveys.

2. Due to the specific forest definition chosen by Liberia, deforestation is a binary process: an

area is either deforested or it is not within a particular time period. However, degradation is

normally treated as a continuous process to some degree or other: for the creation of a

reference level we need to know not only the total area that has been affected by

degradation, but also the degree of this impact. Along with detecting the degradation, the

emissions associated with this degradation are also needed for total degradation emissions to

be estimated. Ideally we would like to know the carbon stock loss before and after a

90 There are now mechanisms being developed to detect in-pixel changes in spectral resolution as a result of a canopy change. See the work of the Carnegie Institution for Science (http://claslite.carnegiescience.edu/en/) and the work of Applied Geosolutions (http://www.appliedgeosolutions.com/)

http://claslite.carnegiescience.edu/en/



127

degradation event, or something that can be related to this loss (for example canopy cover

change).

3. Forests are constantly changing even without human disturbance. Between about 1 and 3 % of large trees die each year in a typical tropical forest, with large tree falls often causing large, natural gaps in the forest canopy. Similarly, trees are constantly growing within forest areas and expanding into current non-forest areas. Thus, not all reductions in canopy cover are anthropogenic, and therefore ‘degradation’, and the natural state for most forests is to increase in biomass through time (so a finding of stability in an analysis may not mean baseline degradation rates are 0). This means that a level of human interpretation or ancillary data is needed to interpret results of changing biomass or canopy cover, and convert them into data on degradation.

In an effort to examine the magnitude of historical degradation in Liberia using the data currently

available, a spatial mapping approach was developed and applied, as explained below. However, it should

be noted that, as in the above section, this approach can only be used to examine the potential prevalence

of degradation and not as a method to estimate degradation activity data.

The landcover map for 2014/15 produced by Geoville/Metria divides the forest class into two useful

classes, one with >80% canopy cover and one with 30-80% canopy cover. If this effort were repeated in

the future the rates of transition between the two canopy cover classes could be proposed as an approach

to estimate a component of degradation. Since in Liberia, there are no non-anthropogenic forest types

with a land cover of 30-80% canopy cover, it can be assumed that any areas in this forest class were either

degraded or are regrowing following deforestation. One approach to estimating at least a portion of

degradation could be to assume areas under the following criteria experienced ‘degradation’:

deforestation was not been detected historically, were in the >80% canopy cover strata in the first land

cover map, and moved to the 30-80% canopy cover in the later map. This approach would still miss a

significant portion of degradation because of the coarse 2-class approach (for example a change from 100-

81% canopy cover, or 70-30%, would be missed) and any degradation that took place but then became

undetectable within the time frame of the two maps. Also the mapping method, using RapidEye and

Landsat data, is imperfect, so there will be some false detections: for example, in reality a pixel could move

from 75 – 78% canopy cover over 2 years, but the first map could incorrectly place it in the >80% canopy

cover class, and the next time in the 30-80% canopy cover, and the pixel could be reported as degraded.

We have no ground-based accuracy assessment of the Geoville/Metria map, so there is no clarity as to

how likely this is to happen. But, given the existing data, this approach is presented below to illustrate the

potential magnitude of coarse degradation.



128

In order to provide a rough estimate of current degradation, we used the Liberia-corrected year 2000

Hansen et al. (2013) and identified areas that were not been deforested between 2000 and 2014.

Comparing these areas to the 2015 Metria/Geoville map provides a rough map of coarse degradation in

Liberia (Figure A1-1). There is no apparent trend in location of potential degradation with respect to

concession boundaries for oil palm, mining, and forest management, or with respect to centers of

population.

Figure A1-1. 'Degradation' estimated by comparing area moving from >80% Canopy Cover to 30-80% canopy cover between

2002-2013



129

This assessment indicated that degradation may have occurred on more than twice as much land area as

deforestation, and emissions from degradation may have accounted for a significant proportion of total

emissions from land use and land use change, depending on the data used to develop emission factors for

degradation. However, to produce reliable estimates of degradation, it would be necessary to have older

maps that have been sufficiently ground-truthed so that transition between forest cover classes can be

known with a high degree of certainty. Because such maps do not exist, the approach used here merely

provides order of magnitude estimates of degradation.

Based on the estimates provided here, it is possible that degradation is a substantial source of emissions.

It is therefore recommended that Liberia include degradation in the REL as a stepwise addition. It is critical

to note that the orders of magnitude estimated vary widely between the two methods described here –

activity-based accounting and land-based or spatial accounting. This further points to the need for

improved data on forest degradation in Liberia. Further details on specific measurement and monitoring

approaches Liberia might adopt for either approach are discussed in below.

Options for Monitoring Degradation in the Future

Many options exist for monitoring degradation, with a wide range of reliability. The accuracy of available

methods is often dependent on the main activities that lead to degradation. It would therefore be useful

for Liberia to conduct a more thorough analysis to identify the primary drivers of degradation in the

country. Following this, a detailed assessment could be done to compare the available methods to monitor

degradation. This should entail identifying the relevance of each option to Liberia’s circumstances, the

required data for each option and the potential to obtain such data, and the costs of monitoring methods.

This appendix describes available options for monitoring degradation, using both land-based and activity-

based approaches.

Land-Based Accounting:

There are four major possible options we see for monitoring degradation in Liberia using the land-based

approach:

Field surveys,

Optical remote sensing data,

Radar remote sensing data, and

LiDAR remote sensing data.



130

Ideally, field data would be used in combination with one of the remote sensing methods, but field data

could be used alone.

The lowest cost option is the use of field data combined with optical remote sensing data. Radar data

combined with field data could provide a higher quality full-coverage option, with no sensitivity to cloud

cover and the potential to reduce field effort with time, but is more expensive. LiDAR data can provide

very high fidelity products, but is unlikely to be cost effective in the short to medium term; it represents

by far the most expensive option.

Details on the four types of data collection, and variants within them, are set out below, as well as

justification for their use and potential problems.

1. Field data

The least capacity-intensive method to monitor forest degradation is through field data collection. Such

data collection could be implemented within an ongoing National Forest Inventory program, which is in

general recommended for Liberia in order to improve estimates of Emission Factors and stocks, reducing

costs further.

Field data for mapping degradation usually takes one of two approaches.

a) Unmarked permanent plots

In this method randomly located plots are set up across study areas (stratified for example by

forest type and accessibility) throughout the country. These plots are inventoried using standard

methods (e.g. the diameter of every tree over 10 cm in diameter measured within a 1 ha square

plot or half hectare circular plot) but are left unmarked: trees are identified by differential GPS

and plot centre/corners are marked in a hidden fashion, for example by hammering iron bars into

the soil to be refound later by metal detectors. These plots can be revisited every 2-3 years and

recensused to estimate changes in carbon stock, and dead trees studied to detect whether their

death was due to human degradation (e.g. cut or fire), or due to natural causes. Training manuals

exist for such methods91, and they have been widely used by voluntary sector REDD+ projects.

Such plots are excellent at ascertaining the rates of tree growth, regeneration, deforestation and

forest degradation throughout a country, producing reliable results. However, in order to produce

narrow confidence intervals, a very large number of such plots are needed – numbering thousands

across the country. Further, in order to produce unbiased estimates of degradation, it is important

91 http://redd.unfccc.int/uploads/2_138_redd_20090302_kyoto_fieldguide.pdf

http://redd.unfccc.int/uploads/2_138_redd_20090302_kyoto_fieldguide.pdf



131

that local communities do not treat the plots any differently to surrounding forest. As it is often

desirable, and sometimes necessary, to involve local communities in setting up and measuring

plots, this in practice can be difficult or impossible.

For the above reasons, it may be that unmarked plots are included as part of a NFI approach, but

not used as the sole method to estimate the rates of degradation, instead being used to validate

other remote-sensing based methods. However, we believe this could potentially represent the

sole approach for Liberia, perhaps on a transitional basis, if well designed and implemented.

b) Stump surveys

Here, trained forests or community members, take a more active approach, roaming through a

landscape on fixed transects and marking the location of any recent treestumps or other signs of

tree clearance. These may be recorded using a field GPS, or perhaps using a mobile phone or tablet

using software as Open Data Kit to record the location and take photographs. The latter method

may produce high quality data, despite requiring less training.

Examples of such surveys producing active data on degradation can be found in the Mapping for

Rights programs92, which work with communities across west/central Africa

Repeated surveys can produce an idea of the rate of removal of trees from an area, and local

increases can quickly be noticed and managed. This approach is often used where active

management, for example to reduce illegal logging, is to be combined with monitoring. It tends to

be much faster acting, but produces data less easily convertible to carbon stocks, than method a).

Like method a), it is maybe best used as a validation method for remote sensing techniques. Unlike

method a) we do not believe it could be used in isolation.

Remote sensing data

Remote sensing data can operate from a satellite, aircraft or unmanned aerial vehicle (UAV or drone). It

broadly consists of three different types, which will be discussed in turn:

92 http://www.mappingforrights.org/

http://www.mappingforrights.org/



132

2. Optical remote sensing data

Optical remote sensing data is the most widely available of the three, with over a hundred satellites

collecting data regularly. There are a number of options for free optical satellite data, with the most useful

for this purpose being:

- Landsat 8, 30 m resolution, revisit every ~14 days

- Sentinel-2, 10 m resolution, revisit every 3-4 days

The resolutions of these free satellites are too coarse to see individual trees, so instead the hope is to

detect degradation through a drop in greenness during the short window after degradation where it is

visible. Also, such satellites may be able to produce classified maps by canopy cover level (the G/M map

was largely based on Landsat 8 data). However, optical data is not ideal because it cannot see below the

top of the canopy, and therefore cannot possibly detect degradation that occurs below the canopy, and

Figure 1 – the three sensor types

1) Optical

2) Radar

3) Lidar

InputBackscatter



133

because cloud-cover often obscures forested areas, sometimes for years at a time. The frequent

observations of new satellites, in particular Sentinel-2, may overcome the cloud cover issue: but the

number of scenes that need to be downloaded and analysed to build up cloud-free data is large and may

raise costs.

Free optical remote sensing data therefore provides the lowest cost means to assess degradation over the

whole country. However, the accuracy of such products may be low: these data are perfect for mapping

deforestation, but far from ideal for degradation.

An alternative could be to use paid-for, higher resolution data. Ideally such data would allow the

discernment of individual canopy trees (for example Worldview-2/3 data with sub-meter resolution). The

G/M mapping used some RapidEye data at 5m resolution, which can discern the canopies of the largest

trees. Such data is suitable for mapping canopy-level degradation with high fidelity, but the data itself is

very expensive, and processing it is also time consuming, and thus it could only be useful as a sampling

tool.

Analysing RapidEye data over a whole country has been undertaken for Guyana as part of their MRV

system, with the particular aim of detecting deforestation and degradation93. This was a high cost option,

both for the purchase of the data and for analysing it. The results are yet to be assessed by a 3rd party, but

appear to be good: however the cost and effort involved was very high, and it is not clear that continued

monitoring at this resolution will be possible, or even necessary.

Metria/Geoville did have success in Liberia in differentiating two different forest classes, 30-80 % canopy

cover and >80% canopy cover, with a training dataset developed using RapidEye (at 5 m resolution) and

the final map based on a combination of RapidEye and medium-resolution Landsat data. Separately, FFI

were able to map degraded forest in Liberia’s proposed reserve Wonegizi using a time series of historical

Landsat data and RapidEye. These approaches have promise, and offer the lowest cost recommended

option for Liberia.

3. Radar remote sensing data

Radar satellite data looks sideways at the world and can penetrate through the forest canopy to obtain

information on forest structure. Therefore radar satellites have been used to map aboveground biomass

93 Bholanath & Cort. 2015. National Scale Monitoring Reporting and Verification of Deforestation and Forest Degradation in Guyana. International Archives of Photogrammetry, Remote Sensing and Spatial Information Sciences, W3: 315-233. http://www.int-arch-photogramm-remote-sens-spatial-inf-sci.net/XL-7-W3/315/2015/

http://www.int-arch-photogramm-remote-sens-spatial-inf-sci.net/XL-7-W3/315/2015/



134

directly, and to map degradation. However, they are limited by a saturation point, typically around 150

Mg ha-1 aboveground biomass for L-band (the longest, and therefore most sensitive, wavelength currently

available), and 50-75 for C-band (which is more widely available).

For Liberia then current radar satellites will not be able to produce continuous biomass maps, as the

maximum AGB in the country far exceeds 150 Mg ha-1. But both L- and C-band provide potential routes

for mapping degradation directly.

Mitchard and colleagues have used L-band data from the ALOS PALSAR satellite to map degradation in

Peru (Joshi et al. 2015) and in Sumatra (Collins & Mitchard, 2015), and have had some success in using it

for monitoring in Sierra Leone in a project in partnership with Winrock. However, data availability is

somewhat limited and the current products from the only L-band satellite orbiting, ALOS-2, are expensive

(around $2000 per 60x70 km scene).

A recent development has been the launch of Sentinel 1a and 1b, two C-band satellites by the EU. These

provide free, regular data. Mitchard is developing a product at the University of Edinburgh called SAREDD

that aims to use C-band data to routinely map forest degradation, and early results are promising: but this

is still very much in a development phase.

Overall radar has potential here. A combination of L-band used to directly monitor biomass and biomass

change in the lower biomass parts of the country, and C-band to monitor degradation in the tall forest

areas (for example looking for logging damage) could provide a high-tech, high accuracy solution for

Liberia. This is presented as a middle cost option.

4. LiDAR remote sensing data

LiDAR data uses laser light looking directly down to give tree height. Repeat surveys can therefore see the

removal of individual trees, and thus it is the only remote sensing method that can guarantee to map

degradation with high accuracy even if the magnitude or size of disturbance is low.

No satellite data is currently collected, so data could only be through aircraft or UAV’s, at high cost. We

are not aware of good aircraft solutions operating in west Africa, and relying on a plane coming from e.g.

South Africa or Europe would be very expensive. On the other hand UAV LiDAR costs are coming down

considerably, with complete solutions from Avion Jaune94 at about $200K, and from Delair Tech95 at about

$400K.

94 http://www.lavionjaune.fr/index_en.html 95 http://www.delair-tech.com/en/packages/dt26x-lidar/features/

http://www.lavionjaune.fr/index_en.html

http://www.delair-tech.com/en/packages/dt26x-lidar/features/



135

Owning such a system would allow Liberia to collect repeat transects on an annual basis, providing

accurate biomass and biomass change maps and thus really understanding the processes of degradation

and regrowth.

Both companies can also collect data on a contract basis for a one-off payment without having to purchase

the planes, and independent service companies such as Carbomap96 can process such data cheaply or

provide training and software solutions

In the long-term LiDAR data could be necessary for a comprehensive, high-fidelity degradation monitoring

system for Liberia – but the cost is likely to be high.

Current LiDAR data collection by palm oil companies such as Sime Derby could reduce the cost: it is

possible the state could share LiDAR costs with plantation development companies, or at least half the

costs of revisit. A LiDAR sampling methodology provides the 3rd, most expensive but highest accuracy,

proposed methods.

Activity-Based Accounting

Approaches for accounting for forest degradation from a range of relevant forest degradation activities in

Liberia are discussed below.

Woodfuel Collection

The analysis of emissions from historical woodfuel collection in Liberia applied to generate initial estimates

as offered above were developed using the Woodfuels Integrated Supply/Demand Overview Mapping

(WISDOM) methodology97 . The estimates were developed based on an analysis conducted by a co-author

of the WISDOM methodology and reflect emissions from the fraction of non-renewable woodfuel harvest.

These estimates were generated by applying a range of regional and global datasets, and can be

considered better than IPCC Tier 1. Improvements to the emission estimates could be realized by

integrating more spatially explicit, country-specific, and more recent data inputs to the WISDOM model

(the estimated emissions listed above represent those for the year 2009).

If Liberia chooses to apply the WISDOM model to account for forest degradation emissions from woodfuel

collection under its REDD+ program, it is recommended that efforts be made to both build in-country

capacity for its application, as well as collect country-specific data to improve the accuracy of its estimates.

96 http://www.carbomap.com/ 97 http://www.wisdomprojects.net/global/

http://www.carbomap.com/

http://www.wisdomprojects.net/global/



136

The WISDOM model can be tailored to fit Liberia’s needs in terms of geographic scope (e.g., Liberia’s

specific administrative units), and consists of modules on demand, supply, integration and woodshed

analysis. Each module requires different competencies and data sources and its contents are determined

by the data available or, to a limited extent, by the data purposively collected to fill critical data gaps.

Information of relevance to wood energy comes from multiple sources, ranging from census data to local

pilot studies or survey data.

The following data sources would improve outputs for each module:

Demand:

Woodfuel demand is largely a function of population and population density, infrastructure, household

energy supply needs, and access to woodsheds. As such, the following sources of data can support the

estimation of woodfuel demand specifically for Liberia:

Population census

Spatial data on infrastructure (e.g., roads, gas pipelines)

Topography

Surveys of household energy needs and use

Supply:

Woodfuel supply is a measure of both the existing biomass in woodsheds as well as their productivity.

Productivity is an important consideration as it accounts for the ability of biomass stocks to regenerate

once harvested for woodfuel use).

The following sources can contribute to the estimation of woodfuel supply:

Biomass stocks (stocks could be tailored to match national forest inventory data)

Productivity (mean annual increment) Integration

Use of spatial data to estimate the demand and supply balance of woodfuel, specific to the desired spatial

resolution. This will identify areas of deficit, surplus, and can help plan for future scenarios.

Woodshed analysis

The analysis for the delineation of woodsheds in Liberia, i.e. supply zones of specific consumption sites

requires additional analytical steps that may be summarized as:



137

Mapping of potential “commercial” woodfuel supplies suitable for urban, peri-urban and rural markets.

Definition of woodsheds, or woodfuel harvesting areas, based on the level of commercial and non-commercial demand, woodfuels production potentials and physical/economic accessibility parameters. Estimation of harvesting sustainability, of woodfuel-related renewable biomass values at subnational level and of woodfuel induce forest degradation rates.

Selective Logging

Under an activity-based approach for estimating emissions from selective timber harvesting, the

accounting methods outlined by Pearson et al. (2014)98 are recommended, whereby data on harvest

volume (activity data) are paired with an emission factor that reflects three emission sources that occur

as a result of logging:

1. emissions from the milling, processing, use and disposal of the felled timber-tree,

2. emissions from incidental damage caused by the timber-tree fall and cutting of the log in the

forest, and

3. emissions from infrastructure associated with removing the timber out of the forest (e.g. skid

trails, logging decks and logging roads).

The method uses the IPCC gain-loss approach, and the total emission factor is the sum of these three

sources of emissions, expressed as units of carbon per cubic meter of timber extracted:

TEF = ELE + LDF + LIF

Where:

TEF = total emission factor resulting from timber harvest (t C m-3)

ELE = extracted log emissions (t C m-3extracted)

LDF = logging damage factor—dead biomass carbon left behind in gap from felled tree and

incidental damage (t C m-3extracted)

98 Pearson T.R.H., S. Brown and F. Casarim. 2014. Carbon Emissions from Tropical Forest Degradation Cause by Logging. Environ. Res. Lett. 9 034017 (11pp). Winrock International. Available at: http://www.winrock.org/sites/default/files/publications/attachments/Pearson%20et%20al%202014%20Logging.pdf

http://www.winrock.org/sites/default/files/publications/attachments/Pearson%20et%20al%202014%20Logging.pdf

http://www.winrock.org/sites/default/files/publications/attachments/Pearson%20et%20al%202014%20Logging.pdf



138

LIF = logging infrastructure factor—dead biomass carbon caused by construction of

infrastructure (t C m-3)

The total emission factor can then be multiplied by activity data derived from timber harvesting statistics,

typically expressed as volume over-bark harvested in cubic meters, to estimate total emissions from

logging operations. Alternately, activity data can be based on area logged, in which case emission factors

must be developed as tons of carbon per hectare. This method is likely to be less accurate as it can be

difficult to identify all logging areas using remote sensing (Indufor 2013).

The data needed to estimate emissions from timber harvesting are given in Table A1-2. Both Tier 2 and

Tier 3 would require original data collection in the REDD+ country. The difference would be in the

completeness of data collection, with a Tier 2 being just a limited sampling of timber harvesting sites to

develop national factors and Tier 3 being more finely stratified by area and by harvesting practices within

the country.

Table A1-2. Requirements and sources of data needed to estimate emissions from timber harvesting

Type of data Specific data needs Sources for Tier 1 data Sources for Tier 2 & 3 data

Activity Data

Timber extraction data (volume per hectare or total volume) on an annual basis

FAO Global Forest Resources Assessment

Government statistics, timber concession reporting, mill reporting

Area of logged forest per year

Limited availability in FAO Global Forest Resources Assessment (often total area of produciton forests only)

Government statistics, timber concession reporting, remote sensing data

Area of logging roads, skid trails, logging decks

Not available

Government statistics, timber concession reporting, high resolution remote sensing data

Emission Factors

Measurements of logged trees (ELE)

Pearson et al (2014) Pearson et al (2014) correlation; Fieldwork/REDD+ NFMS

Extent of incidental damage (LDF)

Pearson et al (2014) Pearson et al (2014) correlation; Fieldwork/REDD+ NFMS

Extent of infrastructure (LIF)

Pearson et al (2014) Fieldwork/REDD+ NFMS



139

Winrock has developed comprehensive, country-specific standard operating procedures and field

inventory approaches for developing emission factors for selective logging that capture the above

emission sources99.

As stated above, activity data reflecting harvested timber volume would need to be available. Official

statistics available may likely only reflect legal timber harvest, thus leading to a significant underestimation

of emissions from timber harvesting due to the persistence of illegal logging in Liberia. As such, Liberia

could endeavour to produce estimates of illegal harvested timber volume through monitoring approaches

or by leveraging data from studies that have been conducted that estimate this value. In the case of illegal

timber harvest, emission factors may need to differ from legal harvest as illegal timber harvesting may be

associated with less associated infrastructure development.

Fire

Emissions resulting from forest degradation due to fire can be generated using global remote sensing

products that are freely available. For example, the MODIS burned area product100 can be used to identify

areas that experienced emissions due to forest fire, although at smaller scales this may overestimate

emissions from fire due to the fact its resolution is coarse (500m). As such, the results produced by

applying this product must be refined not only to exclusively capture only burned area in forests remaining

forests (i.e., degradation, rather than deforestation fires), but it should also be subject to careful

processing to ensure it accurately reflects the magnitude of fires occurring over the period of interest.

An initial analysis of fire using the MODIS burned area product was conducted. This analysis indicated that

fire occurrence and resulting emissions are very low in Liberia, averaging only 225 hectares annually over

the Reference Period. It therefore does not seem necessary that Liberia undertake a more rigorous

method for monitoring fire.

99 Walker, S.M., T.R.H. Pearson, F.M. Casarim, N. Harris, S. Petrova, A. Grais, E. Swails, M. Netzer, K.M. Goslee and S. Brown. 2015. Standard Operating Procedures for Terrestrial Carbon Measurement. Winrock International. 100 http://modis-fire.umd.edu/pages/BurnedArea.php

http://modis-fire.umd.edu/pages/BurnedArea.php



140

Mining

Forest degradation emissions as a result of mining activity can vary significantly, depending on the

magnitude and type of mining. Research conducted by Winrock International in Guyana101 on the impact

of gold mining on the carbon stocks of forests surrounding the mines revealed that forest degradation,

although observed, overall had an insignificant impact on forest carbon stocks, and represented a very

small fraction of overall emissions from deforestation and forest degradation. However, this is likely to

vary by country, and mining may have a more significant impact on forest degradation in Liberia,

depending on the practices. A similar method of assessing degradation from mining activity could be

applied. Under the approach taken in Guyana, mining sites were mapped using high-resolution imagery

and a fixed buffer area outside the mining sites served as activity data. Emission factors were developed

through field data collection. Such an approach should be taken in concert with mapping of other land use

and land use change activities.

Draft Terms of Reference: four potential degradation MRV methods for Liberia

Set out below are terms of reference and rough budgets for conducting a complete MRV cycle for

degradation in Liberia. Costs are given for per cycles of data collection plus processing: obviously at least

two cycles are needed in order to monitor degradation. Cycles could be as far as 5 years apart, but given

the lack of data in this area repeating surveys every 2 years is recommended.

1. Field data only: unmarked field plots 1.1. Background

1.1.1. Degradation and deforestation monitoring can be tracked by repeatedly visiting the same

areas and measuring the diameter and height of all trees in that area

1.1.2. This method is conceptually similar to the Permanent Plot remeasurement protocols used to

assess the growth and mortality rates, and thus calculate carbon sequestration rates, or

intact forest. Therefore, the method can follow broadly the RAINFOR forest plot setup and

analysis protocols102.

1.1.3. However, the accuracy of the method relies on people not treating trees within these plots

differently to normal forest areas. For this reason the plots must have no obvious markings,

101 Brown, Mahmood, Goslee. Unpublished. Degradation around Minded Areas: Methods and Data Analyses for Estimating Emission Factors. Report to the Guyana Forestry Commission. 102 http://www.rainfor.org/upload/ManualsEnglish/RAINFOR_field_manual_version_2016.pdf

http://www.rainfor.org/upload/ManualsEnglish/RAINFOR_field_manual_version_2016.pdf



141

and community members (who will, and should, be involved in monitoring) must agree not

to give away the position of the plots, nor treat them differently.

1.1.4. Though the plots must be unmarked, it is also critical that exactly the same trees are

measured on each occasion the plot is revisited. For that reason permanent markers should

be used (e.g. iron bars hammered fully into the ground), differential GPS points taken, and

photographs collected.

1.2. Number and location of plots

1.2.1. The forest areas of Liberia differ greatly in their risk from deforestation and degradation.

Therefore plots should not be set up with equal likelihood of occurring in any forested area,

as that would require a larger sample size than is necessary. Instead, forest areas should be

stratified based on a) Forest Type and b) Risk Level, and a certain number of plots placed

within each.

1.2.2. We would expect a stratification based on two forest types (low and high biomass) and three

categories of risk (low, medium and high) based on distance to roads/settlements and

protection. Completing this stratification might be within this TOR, or may be separately

completed by a different group.

1.2.3. The number of plots within each strata can be determined based on the area of each strata

and expected standard deviation of change responses, following standard CDM

methodologies103. Given the lack of data this would need to be estimated with consultation

with experts. It is expected that the total number of plots to be set up across Liberia would

be at least a thousand, with a much higher density of plots in the high risk than the low risk

areas.

1.2.4. Plots should be placed randomly within each strata. Using a systematic grid is not

recommended for this purpose as the distances between each plot would be unchanging,

preventing the calculation of statistics as to the spatial correlation of degradation activities.

1.3. Plot measurements

1.3.1. We recommend setting up plots are square and 1 ha in size. This large size reduces the

edge:volume ratio, minimising errors caused by plot misplacement on uncertainty over

whether trees are ‘in’ or ‘out’ of the plot. Such plots are also less sensitive to the removal or

death of a single tree.

1.3.2. Plot location should be given by a random number generation within a strata, using software

such as QGIS or ArcMap. However once in the field the plot can be moved slightly (<10 m)

from these coordinates to avoid dangerous obstacles such as rivers or cliffs being within the

plot. Plots should only be set up if they are clearly ‘forest’ within Liberia’s definition.

1.3.3. Actual plot corners should be recorded using a differential GPS, and iron bars (rebar)

hammered into the ground at the plot corners.

103 http://cdm.unfccc.int/methodologies/ARmethodologies/tools/ar-am-tool-03-v2.1.0.pdf

http://cdm.unfccc.int/methodologies/ARmethodologies/tools/ar-am-tool-03-v2.1.0.pdf



142

1.3.4. Within each plot all trees with a diameter at breast height greater than 10 cm should have

their diameter and height measured, and their species identified. RAINFOR tree census

protocols should be followed, but by necessity the Point of Measurement cannot be marked

and tree tags cannot be used. Trees should however all be given a unique number and their

position mapped, as normal with RAINFOR plots.

1.3.5. Hemispherical photos should be captured on every 20x20 m vertex, in order to allow the

plot to be repositioned and the fate of trees to be tracked.

1.3.6. During a repeat survey, every effort should be made to re-find trees recorded in the

previous survey. If they have died, it should be recorded as far as possible whether the death

was natural or due to anthropogenic harvesting.

1.3.7. If is it seen as significant, trees with a diameter between 1-10 cm could be assessed within

three 20 x 20 m subplots.

1.3.8. If desired, soil cores or soil pits could be dug in a subset of plots in order to assess changes in

below-ground biomass. However, this may be best left to the NFI as this could make the

location of the plots obvious.

1.4. Biomass calculations

1.4.1. Aboveground biomass should be calculated using the Chave et al. (2014) equation104, unless

a Liberia-specific equation is developed. Wood density values can come from the Global

Wood Density Database105 on a species basis, or local values used if available.

1.4.2. After a repeat survey, the change in biomass should be separately attributed between tree

growth, tree death (natural), and tree death (anthropogenic).

1.4.3. These numbers can then be scaled, using the strata, to the whole of the country. Ideally the

strata would be reassessed, using new remote sensing data, at every census.

1.5. Indicative costs

1.5.1. The stratification work would require field data collection throughout the country

(potentially as part of a first NFI census), the analysis of remote sensing data, and modelling

of degradation risk. These activities are likely to take place anyway, so may not be addition

to the degradation work. But we would estimate costs of $300K for 3 months field work by 6

teams, and 60 high-level consultant days to analyse these data, Landsat satellite data, and

produce a stratified map. However, on the first occasion, it may be that the M/G map and

spatial layers of roads and settlements could be used to produce a lower quality

stratification for approximately 20 consultant days, but this will necessitate the collection of

more plot data.

1.5.2. Setting up the field protocols, creating the random points, and conducting training would

likely involve on the order of 50 days of a high-level consultant.

104 http://onlinelibrary.wiley.com/doi/10.1111/gcb.12629/abstract 105 http://datadryad.org/handle/10255/dryad.235

http://onlinelibrary.wiley.com/doi/10.1111/gcb.12629/abstract

http://datadryad.org/handle/10255/dryad.235



143

1.5.3. Each plot would take 2-5 days for a standard team of 4 field operatives and 1 field

scientist/manger to complete. The time taken would increase dependent on the density of

forest within the plot. Travel time between plots could be at least 1 day, and the team would

need a rest, so an assumption of 1 plot per team per week would be reasonable. Previous

fieldwork in Gabon has suggested an overall cost of about $2500/week as a minimum for

logistical support, equipment and salary for such a team, but costs in Liberia could be higher

or lower depending on how many high capacity local people could be hired.

1.5.4. Assuming 1000 plots were to be set up an indicative cost of US$2.5M would be needed for

the field data collection per census.

1.5.5. After each census a team would need to enter data, validate the data, and calculate the

carbon storage. This is likely to involve 1 skilled day’s work per plot.

1.5.6. A final report after the second census would integrate all the plot data and provide change

statistics and error estimates, involving perhaps 50 days of a high-level consultant’s time

with the support of 2 local staff involved in the data entry and validation process.

1.5.7. Total cost is thus very dependent on plot numbers chosen, but could be well over $3M.

2. Low tech: Optical data + Field Data. 2.1. Background

2.1.1. Metria/Geoville and FFI have already shown the capacity for a combination of Landsat and

RapidEye data to classify forest by different levels of canopy cover.

2.1.2. In this method Landsat 8 or Sentinel 2 data would be used to map forest canopy cover into 3

broad classes: 30-50%, 50-80%, and >80%. The additional class over the G/M map allows for

higher fidelity degradation and regrowth monitoring.

2.1.3. Training and validation data would be provided by a combination of field plots and RapidEye

data

2.2. Field plots

2.2.1. Ideally no additional field costs would be involved in setting up this program. As part of an

NFI we hope that Liberia will be assessing the biomass and canopy cover of a large number

of plots across the country. These could be used to create canopy cover maps from RapidEye

satellite data, and to assign biomass values to the different canopy cover classes.

2.2.2. If no NFI were to take place, then hundreds of field plots within each strata (divided by risk

and canopy cover) would need to be set up, as per 1. Above. A lower total number of plots

would be needed, but an additional parameter, canopy cover, would need to be collected

using either hemispherical photographs or an LAI2000.

2.2.3. After two assessments had taken place using this method, a Stump Assessment fieldwork

campaign is recommended, where areas that have changed from >80% canopy cover to

lower canopy cover classes are randomly visited and it is confirmed whether tree stumps are



144

present. This would allow an independent verification of the method, but is not essential as

other data (such as an NFI or LiDAR campaigns) could provide alternative verification.

2.3. Remote sensing data ordering

2.3.1. Either Sentinel 2 or Landsat 8 data could be used for this task. We would recommend

Sentinel 2 as it is better guaranteed to be available in the future (2 satellites are guaranteed

to be orbiting into the 2030’s), whereas Landsat relies on a single satellite, and it has a 10 m

resolution rather than a 30 m resolution. Further, it has a higher repeat rate guaranteeing

more cloud-free images.

2.3.2. A single, cloud-free mosaic for a single year for the country should be created using the data

in 2.3.1, ideally using data from a ~1 month period at the height of the dry season.

2.3.3. Data at a higher resolution is essential to link the field data to the 10-30 m resolution

satellite data. RapidEye provides a reasonable balance between resolution (5 m) and cost –

for large areas indicative costs are around or below $1.5/km2 for an enterprise licence,

though processing costs far exceed that. We would recommend ordering RapidEye data for

3-5 areas distributed across the country, covering a total of perhaps 10,000 km2, a little

under 10% of the country.

2.4. Remote sensing data processing

2.4.1. The RapidEye data should be used to make a map of the three canopy cover classes based

on the field plot data, using a machine learning technique suitable for classification such as

RandomForest, at a 10 m resolution, including texture metrics developed from the high

resolution data. This should have an accuracy against 50 % of field plots held back for testing

automatically by a normal RandomForest implementation of >95 %.

2.4.2. These high resolution maps should then be used to train and test a country-wide map using

the Sentinel-2 data. If Landsat is used, they should first be degraded to 30 m. Again, a

technique such as RandomForest would be preferred for dealing with these large training

and testing datasets, though if that is computationally too challenging a technique such as a

Support Vector Machine could be implemented. This map should have an accuracy >90 %.


2.5.1. Landsat and Sentinel-2 data are free, but a system to download and store the data in Liberia

could involve a hardware cost of $20-30K, plus annual maintenance costs of at least $10K.

Alternatively an external consultant could do the remote sensing data analysis, but that

might be politically unsatisfactory for Liberia.

2.5.2. RapidEye data purchase cost is expected to be on the order of $15K for 10,000 km2.

2.5.3. Analysing the data would probably involve a team of 3 GIS technicians working for 80 days

each, under the direction of 1 remote sensing specialist working for 50 days, at a total cost

of approximately $170,000.

2.5.4. Total cost is thus $220,000 per survey.



145

3. Medium tech: Radar data 3.1. Background

3.1.1. L-band radar data from has been shown to be effective for mapping biomass change directly

up to 150 Mg/ha biomass106, which we estimate represents about half of Liberia’s forest

area and probably 2/3 of its area of forest change.

3.1.2. Separately, experimental work with C-band radar suggests that degradation an deforestation

in intact forest, above the saturation point for L-band radar, can be detected reliably using

an algorithm developed at the University of Edinburgh called SAREDD

3.1.3. L-band radar data is only collected by the JAXA satellite ALOS-2 PALSAR-2. This data is

charged, at a cost of approximately $2000/scene. Covering Liberia requires 34 scenes:

3.1.4. C-band radar is collected by a number of satellites, but since 2015 free data has been

available from the Sentinel-1 satellite series on an approximately monthly basis.

3.1.5. Analysing radar data is computationally difficult and involves specialist skills unlikely to be

easily developed in Liberia. Therefore these methods would probably need to be applied,

initially at least, outside Liberia by external consultants.

3.2. Field data

3.2.1. As with method 2., field data would need to be collected if not already collected as part of

an NFI. This would be for the lower biomass regions in order to locally calibrate a L-band

106 http://onlinelibrary.wiley.com/doi/10.1111/j.1365-2486.2011.02551.x/full http://www.tandfonline.com/doi/abs/10.1080/17550874.2012.695814 http://www.biogeosciences.net/12/6637/2015/

http://onlinelibrary.wiley.com/doi/10.1111/j.1365-2486.2011.02551.x/full

http://www.tandfonline.com/doi/abs/10.1080/17550874.2012.695814

http://www.biogeosciences.net/12/6637/2015/



146

radar-biomass relationship. At least 300 biomass plots would be needed, distributed

throughout the lower canopy cover strata. No canopy cover data would need to be

collected.

3.2.2. No field data would have to be collected for the C-band SAREDD system, though validation

data in the form of stump surveys would be useful for validation purposes.

3.3. Satellite data processing

3.3.1. Level 1 ALOS-2 PALSAR-2 scenes would be downloaded for the year of interest. The scenes

would ideally be from a single season, ideally the dry season, collected within a month of

each other.

3.3.2. These would be processed to remove terrain effects at a 25 m resolution, and a seamless

mosaic created.

3.3.3. Through comparison to the ground data a biomass map, with pixel-level uncertainty map,

would be created. A saturation point of the method at approximately 150 Mg/ha is

anticipated, but it could be higher or lower based on this analysis.

3.3.4. Differencing of data below this threshold would produce maps of the magnitude of forest

loss, in tonnes biomass.

3.3.5. The SAREDD system relies on frequent observations, so every Sentinel-1 scene collected

over Liberia would be downloaded and processed into calibrated and filtered stacks at a 50

m resolution.

3.3.6. Areas of low biomass forest would be removed using the ALOS-2 map (the Sentinel-1

method is thought not suited to data with a biomass lower than 100 Mg/ha). The SAREDD

algorithm would then be run over data with a higher resolution, producing output maps of

degradation through time.


3.4.1. ALOS-2 PALSAR-2 data costs would be $68,000 at catalogue prices per survey, though a

volume discount might be negotiable.

3.4.2. ALOS-2 PALSAR-2 processing costs (assuming field data has been collected and provided

suitably) are estimated at 50 high-level consultant days plus 100 GIS technician days per

survey, giving a total processing cost of approximately $125K.

3.4.3. Sentinel-1 data is free, but the SAREDD team at the University of Edinburgh is likely to

charge approximately $5/km2 to provide maps giving the date of degradation or

deforestation for a year. Assuming 60,000 km2 (about half) of Liberia were covered that

would produce an indicative monitoring cost of $300,000.

3.4.4. Total cost is thus $493K/survey

4. High tech: LiDAR Data 4.1. Background

4.1.1. LiDAR provides the highest fidelity method for mapping biomass



147

4.1.2. LiDAR data collected over a stratified random sample of 10 % of the country could,

combined with a robust stratification conducted elsewhere, provide robust estimates of

degradation

4.1.3. Costs are likely to be higher than under the other methods, but this Tier-3 style monitoring

system would have higher accuracy.

4.2. Field data

4.2.1. Field data collection would be necessary under the LiDAR study areas. We would

recommend at least 300 plots in total

4.3. LiDAR data collection

4.3.1. Highest fidelity maps would be produced from using survey-grade LiDAR with a high point

density (at least 5 points per m2). Full waveform data collection should be considered (e.g. as

offered by a combination of Reigl and Carbomap), but a system collecting multiple returns

per point should be sufficient.

4.3.2. We recommend LiDAR is collected over a total of 10 % of the country, disturbed among 20

blocks. These would be stratified by forest type and threat.

4.3.3. LiDAR could be collected using either an aircraft or, at potentially lower cost though using a

developing technology, a UAV

4.4. LiDAR data analysis

4.4.1. The LiDAR data should be processed to give layers such as Top Canopy Height, Mean Canopy

Height, and canopy cover. These can then be related to biomass from the field plots using

methods such as those described by Asner et al.107.

4.4.2. Continuous maps of biomass at a high resolution (20 m) and low, known error, can then be

compared from census to census.


4.5.1. Indicative costs are hard to give for LiDAR as this is a rapidly advancing technology with costs

reducing. But costs are likely to be in the order of $10/ha for data collection and $5/ha for

processing. Therefore conducting such a survey over ten percent of Liberia, which is

1,100,000 ha, would be approximately US$16.5 million.

107 http://ghislain.vieilledent.free.fr/Wordpress/wp-content/Asner2012-Oecologia.pdf

http://ghislain.vieilledent.free.fr/Wordpress/wp-content/Asner2012-Oecologia.pdf



148

ANNEX 2: ECONOMIC ANALYSIS OF BASELINE DRIVERS OF

LAND USE CHANGE IN LIBERIA

Developed by Brent Sohngen (Sylvan Acres Limited Liability Company), Sukwon Choi, and Shelby Stults Liberia has a relatively large area of forests and due to national circumstances they have not recently been heavily harvested or deforested. Using developed datasets and models, an analysis was conducted on potential deforestation and degradation drivers to project potential future deforestation and timber harvesting in Liberia. The analysis also addressed future mining activity and palm oil production for Liberia. Focusing on these three critical sectors, forestry, mining, and palm oil, the authors have been able to locate

and use data to conduct statistical analysis on each of the sectors. Importantly, the results should be

interpreted with some caution given the sparsity of the data available from Liberia. Data on the timber

and palm oil outputs and prices are derived from the UN FAO (2016). These data are collected from local

sources, and then are reported to FAO from the government. Data on economic indicators were obtained

from the World Bank and other sources. While FAO and the World Bank are excellent sources of data in

general, most of the data must originally be collected in Liberia. Given the long running civil conflict, the

completeness and uncertainty of the data is unknown. The data on mining outputs was obtained from

the British Geological Survey (2016). These data were obtained mainly through government reports and

also information on exports over the years.

The report begins by presenting the model developed for the forestry sector. This model consists of a

supply and demand system, with prices based on export prices. A supply and demand model is estimated

because there appears to be a viable local market for wood. Unfortunately we do not have a time series

for domestic wood prices, but we use export prices as a proxy. The report then presents the models

developed for the mining sector, focusing on gold, diamonds and iron. Only the supply side of these

models is estimated, since the local demand for these minerals is limited and we assume prices are

exogenous. These three minerals appear to be the most important sectors for mining in Liberia, although

cement has recently become a larger industry. The final section discusses the results for the palm oil

sector, for which we also develop a supply and demand model. Given estimates of export quantities, it

appears that the largest demand for palm sourced form Liberia is local, so we model a supply and demand

system.

The authors also attempted to correlate prices in these various sectors to actual land use change that

occurred from 2001 to 2014. This analysis, however, was not successful. We did not test the effect of



149

agricultural prices in this analysis. To better correlate the individual sectors with land use change, a more

thorough analysis of what has caused the land use change and where it occurs would be necessary. Aside

from general infrastructure development that cannot be attributed to any one sector, there appear to be

three main drivers of deforestation in Liberia, agriculture, palm oil development, and mining. Spatial

analysis with land use change data could be conducted to determine which of these drivers are primarily

responsible for land use change. If other factors like existing development, existing roads, population

centers and whatnot are controlled, then it is likely that the land use change can be correlated with prices

in markets and a more quantitative assessment of potential future land use changes take place.

Forestry Sector Model

This analysis projects future forest harvesting in Liberia based on an estimation using data from the period

1961 to 2014. The projection is based on modeling a demand and supply system for wood products in

Liberia. Data on timber production, exports, and price variables are obtained from FAO FOREST-STAT

database (UN FAO, 2016). Other data on income and other factors was obtained from the World Bank

(2016).

Figures A2-1 and A2-2 show industrial roundwood production in Liberia, as well as production in several neighboring countries. Total production and export of industrial round wood in Liberia move together between 1961 and 2003 because roundwood harvests have historically been driven by the export market in Liberia. There is a strong cyclical pattern to timber harvests in Liberia related partially to fluctuations in global prices and partly to policy changes in Liberia, such as civil conflict. During the civil war between 1980 and 2003, the entire economy in Liberia was devastated and the GDP per capita in 2003 dropped to the 17% of its pre-war level in 1979 (World Bank, 2016). In the same period, the production and export of timber increased and then abruptly fell in 1995, recovering during the latter part of the 1990s. During the civil war, revenue from forest sector was linked to the illegal arms trade (Blundell et al, 2003). The UN Security Council placed sanctions on exports between 2003 and 2006, and there was no timber export in that period. While there is limited detailed data and information on Liberian forest (Halton, 2013), there was wide spread corruption in forest sector during the conflict era. After a review of forest concessions in 2004, it turned out that the total forest concession area was 2.5 times the total forest area in Liberia and 100% of timber companies violated the laws (Blundell et al, 2005). In 2006, after the sanctions were lifted, previous concessions were nullified and new concessions were granted (Blundell et al, 2007). The period after 2007 appears to have significantly more stable timber harvesting than the period before, and timber harvesting appears to be on an upward trajectory.



150

Figure A2-1. Industrial Roundwood Production and Exports in Liberia (UN FAO, 2016)

Figure A2-2. Total Industrial Timber Production in Liberia and several adjacent countries (UN FAO, 2016)



151

Methods

To make future projections, we estimate a supply and demand system. The supply and demand system

can then be used to project future outputs based on assumptions about future trajectories in exogenous

variables. The supply and demand system we estimate is given as:

Supply: Ln(QSt )= α1 + α2Ln(Pt) + α3DSt + α4Ln(El_Nino) + α5Ln(labor) +et

Demand: Ln(QDt )= β 1 + β 2Ln(Pt) + β 3Ln(GDPPCLBt) + β 4Ln(GDPPCEUt) + β 5Ln(ImpPEUt) + β6Ln(Cote_Pt) + β7DC+ et

Equilibrium: Ln(QSt )= Ln(QDt)

Where:

Ln(QSt ): Natural log of timber supplied in cubic meters (UN FAO, 2016) Ln(Pt): Natural log of timber export price (FUN AO,2016) Ln(labort): Natural log of population aged 15 – 64 (World Bank, 2016) Ln(El_Nino): logged index of the strength of El Nino (NOAA, 2016) DSt: Dummy variable; 1 if during the period of UN sanctions (2003-2006), 0 otherwise. Ln(QSt ): Natural log of timber demanded in cubic meters (FAO, 2016) Ln(GDPPCLBt): Natural log of Gross Domestic Product per capita in Liberia (World Bank, 2016) Ln(GDPPCEUt): Natural log of Gross Domestic Product per capita in EU (World Bank, 2016) Ln(ImpPEUt): Natural log of import price in EU (UN FAO, 2016) Ln(Cote_Pt): Natural log of timber export price in Cote d'Ivoire (UN FAO,2016) DC: Dummy variable for the period of civil conflict (1980-2003).

The supply side of this model is composed of timber export prices, an index for El Nino, the supply of labor, as measured by population, and a dummy variable for the period of sanctions. The parameter on prices is expected to be positive, as a higher price will induce more output. The El Nino index controls for climatic fluctuations that will affect weather patterns generally and influence the supply of wood from forests. Labor accounts for the supply of labor and should be positively correlated with supply. Sanctions are included as a supply side variable because they reduce investments and efforts to produce wood. The demand side is composed of export prices, income in Liberia (GDP per capita), income in the European Union, import prices in the European Union, export prices from the Ivory Coast, and a dummy variable for the period of conflict, or civil war. It is expected that the parameter on prices will be negative, and the parameters on income will be positive. The sign on import prices in the EU should be positive, but the sign



152

on export prices from Ivory Coast will be positive or negative. We anticipate that the sign on civil conflict will be positive given the results Blundell et al. (2005).

Results

The results of the empirical estimation are shown in Table A2-1. On the supply side, export prices have a positive sign, as expected, suggesting that higher prices for export logs increase supply. Sanctions have a negative effect, but are not significant. El Nino is also insignificant. Labor supply is negative and significant. This is surprising, but if an increasing labor supply is also available for competing activities, which could reduce the supply of wood. In the demand system, the price has a negative sign as expected. Income in GDP and income in the EU both have a positive impact upon demand. Import prices in the EU are not significant, although export prices in the Ivory Coast are positive and significant. Thus, if prices rise in the Ivory Coast, then demand for wood increases in Liberia. The conflict dummy variable is positive and significant suggesting that demand for wood was generally higher during the conflict. Table A2-1. Estimation results of supply and demand in Liberia (n=54, 1961-2014)

Parameter Description Estimate Std. Error

α1 Intercept (S) 41.39*** 9.29

α2 Export Price (log) 1.30*** 0.29

α3 Sanction -0.12 0.34

α4 El Nino 0.00 0.01

α5 Labor (log) -1.83*** 0.62

β1 Intercept (D) -63.44** 28.18

β2 Export Price (log) -3.67** 1.52

β3 Liberia GDP per capita (log) 1.33*** 0.31

β4 EU GDP per capita (log) 6.59** 2.51

β5 Import Price in EU (log) 0.7 0.74

β6 Export Price in Ivory Coast (log) 1.54** 0.64

Β7 Conflict Dummy 1.7*** 0.52

Historical harvests and the projection of future harvests through 2025 are shown in Figure A2-3. To make the future projections, we make the following assumptions about the exogenous variables:

Labor supply: Rises at 1% per year GDP per capita in Liberia: Constant GDP per capita in Europe: Rises at 2% per year



153

Import prices in EU: Rise at 2% per year Export prices from Ivory Coast: Rise at 2% per year

Prices and quantities are endogenous, so we predict an equilibrium price and quantity for each historical year with the model and compare it to the actual data. Then we predict the future for a decade with the model and the assumptions above. Based on these estimates, our model suggests that output will increase 3.3% per year. The output data are provided from 1990-2025 in Table A2-2. The results of the future projection suggest that output falls in 2014, mainly due to the reduction in export prices in the Ivory Coast. We assume that export prices in Ivory Coast begin rising again after 2014, leading to higher projected harvesting in Liberia as well.

Figure A2-3. Historical actual, historical predicted, and future predicted timber harvests in Liberia. Historical actual data based

on UN FAO (2016)

0

200000

400000

600000

800000

1000000

1200000

1400000

1600000

1800000

1961 1971 1981 1991 2001 2011 2021

cub

ic m

eter

s

Data Prediction



154

Table A2-2. Timber outputs in Liberia, 1990-2025. Data is actual data from FAO and used as input in the model. Prediction are

the predicted values from the model estimated and presented in Figure A2-3

Data Prediction

Year m3/yr 1990 1,128,049 959,929

1991 746,985 945,686

1992 1,034,988 850,380

1993 934,998 622,852

1994 629,008 514,002

1995 228,000 557,869

1996 163,996 497,453

1997 222,994 507,100

1998 321,001 549,838

1999 516,020 538,622

2000 1,114,036 646,400

2001 1,162,054 684,527

2002 1,544,020 725,047

2003 979,992 652,690

2004 329,984 435,864

2005 329,984 413,200

2006 360,015 444,350

2007 360,015 534,942

2008 419,996 601,187

2009 419,996 460,381

2010 479,980 471,161

2011 483,981 502,351

2012 517,984 472,020

2013 517,984 500,999

2014 517,984 401,663

2015 415,583

2016 430,082

2017 445,087

2018 459,734

2019 474,863

2020 490,491

2021 506,632

2022 523,305



155

2023 540,527

2024 558,315

2025 576,689

Mining Model

Data on mineral outputs for Liberia and neighboring countries were obtained from the British Geological Survey (2016), which has kept track of data on mining outputs from most countries around the world. The data for Liberia and other countries, while stored in the BGS database, are composed of numerous estimates of outputs given that data were not routinely kept in Liberia. The data for Liberia indicates that cement, diamonds, iron ore are three major commodities produced in Liberia. Gold also is produced. We have not been able to develop a statistically valid model for cement. We have been able to develop models for the other three commodities. We did not conduct analysis of the spatial location of mines, although in our search for data, we did find several datasets that provide information on where current mines are located in Liberia. The quality of this data, however, is unknown, and it is not clear if historical data are available. We also do not have data on individual mine output. To our knowledge, this type of data is not available publicly as it usually is kept privately by mine owners or operators, unless they are otherwise required to report outputs to the government. Data on prices were obtained from the US Geological Survey (2014), and data on other economic factors in the countries of interest were obtained from the World Bank (2016). Of the three mining sectors we analyze, iron ore and diamond mining appear to be the most important commercially by volume and value. To determine how various factors affect mining output in Liberia, we construct a mining supply function for each of the commodities. The main factors expected to influence supply are prices for the commodities, labor supply in Liberia (measured by population), exchange rates, and other factors.

Gold

Starting with gold, the supply function we estimate is given as

Ln(Qt) = β1 + β2ln(Pt-1) + β3Golddt + β4Goldd2t + et (1)

Where,



156

Qt = output of gold (ounces per year)

Pt-1 = lagged real price of gold in US dollars, deflated to 1998 real US$

GoldDt = 1 if year ranges from 1999 to 2006, 0 otherwise.

GoldD2t = 1 if year ranges from 2007 to 2014, 0 otherwise.

For this model, we explain gold output as a function of gold prices and two dummy variables representing

different periods. Prices have the most important impact on output, but it is clear from the data that the

period 1999 to 2006 experienced very low gold output, due to civil war and sanctions. We control for the

increase in output that occurred after 2006 with a dummy variable for the period 2007-2014.

Table A2-3. Liberia Gold Supply Model, estimated from historical data

Parameter Description Estimate Standard

Error

β1 Intercept -39.596* 21.564

β2 lngoldpricereallag 2.674** 1.309

β3 Golddummy -3.089** 1.437

β4 golddummy2 0.27 1.548

** Significant at 0.05 level, * significant at the 0.10 level The parameters in the model generally make sense (G3). The price of gold is positive and significantly different from 0. Higher prices imply increased gold output. The gold dummy variable is large, negative and significant, as expected. This parameter controls for the lower level of gold extraction that occurred during this period. The gold dummy for the period after 2006 is positive, but not significant. Prices have a large impact on rising outputs during this time period in this model. The historical actual output, historical predicted output based on the model above, and the predicted future output from the model is shown in Figure A2-4 and the output data are provided in Table A2-4. The model predicts relatively stable outputs in the future, in line with the assumption that gold prices remain fairly stable. Should gold prices rise, these outputs would be expected to increase. On the other hand if prices fall, our model would predict that outputs will decline.



157

Figure A2-4. Historical output and predictions for the gold model for Liberia

Table A2-4. Gold mining outputs in Liberia, 1990-2025. Data is actual data from British Geological Survey and used as input in

the model. Prediction are the predicted values from the model estimated and presented in Figure A2-4

Data Prediction

Year Kilograms/yr

1990 600 139

1991 600 124

1992 700 94

1993 700 81

1994 0 83

1995 800 13

1996 700 268

1997 500 79

1998 800 47

1999 25 2

2000 25 1

2001 0 1

2002 0 0

2003 0 1

2004 110 1

0

200

400

600

800

1000

1200

1400

1600

1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 2020 2025

Kilo

gram

s

Gold Data Gold Predicted



158

2005 25 7

2006 9 2

2007 311 145

2008 624 200

2009 524 373

2010 666 428

2011 449 807

2012 641 1,226

2013 600 1,448

2014 535 881

2015 688

2016 734

2017 699

2018 688

2019 690

2020 691

2021 691

2022 691

2023 691

2024 691

2025 691

Iron Ore

The second model for Liberia is the Iron Ore model. This model is estimated as

Ln(Qt) = β1 + β2ln(Pt-1) + β3ln(Popt-1) + β4ln(ERt-1) + β5IronDt +

β6YY11 + β7YY12 + β8YY13 + β9YY14 +et

Where,



ERr-1 = exchange rate lagged one year (Official Local Currency versus US $)

Popt-1 = Population from 15 to 64 years lagged one year

IronDt = 1 if year ranges from 1993 to 2010, 0 otherwise.



159

YY11 = 1 if year =2011

YY12 = 1 if year =2012

YY13 = 1 if year =2013

YY14 = 1 if year =2014

We use the same variables for the iron ore model in Liberia, except the dummy variable in this case is for the years 1993-2010. During this period, iron ore extraction was 0 based on the data we have obtained. Iron ore extraction recovered in 2011 and rose through 2014, but growth in output appears to have slowed over the last 4 years, so we include individual year dummy variables to account for this trend. The results of the model are shown in Table A2-5. The parameter on price is positive and significantly different from 0 as expected. The population parameter is negative and significant. This is not expected, but the overall trend in iron ore extraction over the time period has been fairly flat, while population has been increasing, so the parameter estimate makes sense. The 1993 to 2010 dummy variable is negative and highly significant as expected. The individual year dummy for 2011 is also negative and significant. The reason for this is that 2011 output was lower than the average output for the years 1970-1992. Output however increased after 2011 in years 2012, 2013 and 2014 relative to the period before 1993. The individual year dummies for 2012-2014 are not significantly different from 0 indicating that we cannot distinguish output in these years differently from the period before 1993.



160

Table A2-5. Liberia Iron Ore Supply Model


Error

β1 Intercept 28.741*** 5.95

β2 Lnironpreallag 0.846** 0.338

β3 lntpop1564lilag -2.306*** 0.812

β4 iron0li -19.402*** 0.244

Β5 Offexchlilag 0.001 0.006

Β6 yy11 -1.833*** 0.373

Β7 yy12 0.078 0.525

Β8 yy13 0.84 0.64

Β9 yy14 0.926 0.718

*** Significant at 0.01 level ** significant at the 0.05 level. The historical actual iron ore output, historical predicted iron ore output (predicted based on model), and

the future predicted iron ore output for Liberia is shown in Figure A2-5 and the output data are provided

in Table A2-6. Output falls from the 1970s through the 1980s, largely because prices for iron ore were

falling. The civil war in Liberia likely also played a role, although when we test for the effects of the civil

war explicitly, the effects are not significant. Output falls completely to 0 in 1993, as noted before and

remains at 0 through 2010, when iron ore mining commences again in Liberia.

The future prediction is made assuming that iron ore prices remain constant in the future, population rises

at the same rate as used in the gold model above, and exchange rate remain constant. The rising trend in

output is obtained by adjusting the YY14 data in the future to increase at 0.10 per year. This introduces a

modest increase trend in production, which follows the increasing trend from 2011 to 2014, but at a slower

rate.



161

Figure A2-5 Historical iron ore production, historical predicted iron ore production and future predicted iron ore production.

All predictions based on the model in G5 above

Table A2-6. Iron mining outputs in Liberia, 1990-2025. Data is actual data from British Geological Survey and used as input in

the model. Prediction are the predicted values from the model estimated and presented in Figure A2-5

Data Prediction

Year Tons/yr

1990 3,981,000 7,108,469

1991 1,200,000 6,494,878

1992 1,710,000 6,321,325

1993 0 0

1994 0 0

1995 0 0

1996 0 0

1997 0 0

1998 0 0

1999 0 0

0

5000000

10000000

15000000

20000000

25000000

30000000

1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 2020 2025

Ton

s

Iron Data Iron Predicted



162

2000 0 0

2001 0 0

2002 0 0

2003 0 0

2004 0 0

2005 0 0

2006 0 0

2007 0 0

2008 0 0

2009 0 0

2010 0 0

2011 386,968 423,907

2012 2,369,850 2,606,485

2013 4,698,281 5,074,744

2014 4,813,676 5,078,190

2015 4,988,790

2016 5,204,505

2017 5,429,549

2018 5,664,323

2019 5,909,248

2020 6,164,764

2021 6,431,329

2022 6,709,420

2023 6,999,536

2024 7,302,196

2025 7,617,943



163

Diamonds

The diamond model in Liberia is given as

Ln(Qt) = β1 + β2ln(Pt-1) + β3ln(Popt-1) + β4SanDt + β5ConDt +et (4)

Where,



Popt-1 = Population from 15 to 64 years lagged one year

SanDt = 1 if year ranges from 2003 to 2006, 0 otherwise (sanctions)

ConDt = 1 if year ranges from 1980 to 2003, 0 otherwise (conflict)

The conflict period is the period over which Liberia was in civil conflict, which lasted from 1980 to 1993.

Sanctions were imposed in 2003 and lasted until 2006. We include dummy variables to control for both

of these. The price parameter is positive and significant, indicating that diamond mining is positively

related to diamond prices (Table A2-7). Population in this case is positively correlated with diamond

mining. The conflict period has a positive correlation with diamond output while sanctions have a negative

correlation. This makes sense and suggests that the sanctions had the intended impact on diamond

outputs. We do not include exchange rates here because they do not have a significant impact and the

sanction and conflict variables appear to capture fairly important supply impacts.

Table A2-7. Diamond Supply Model


Error

β1 Intercept -37.81*** 8.427

β2 Real diamond price (logged and lagged) 0.656*** 0.065

β3 Population (logged and lagged) 0.734*** 0.143

β4 Sanction Dummy -1.321*** 0.196

β5 Conflict dummy 0.949*** 0.141

*** Significant at 0.001 level



164

The predicted output of diamonds for Liberia is shown in Figure A2-6 and a table of the output data is provided in Table A2-8. Diamond production fell from relatively high levels in the early 1980s to very low levels during the period of sanctions from 2003 to 2006. They have recovered since then. The production level in 2014 represented a significant increase in production which our model does not capture. This increase may have resulted from the opening of a new mine. The prediction from 2015-2025 suggests that mining increases slowly. We have assumed that diamond prices remain constant over this time period, and that population increases.

Figure A2-6. Historical diamond production, predicted historical diamond production and future

The three mining models suggest that mining outputs have recovered in Liberia after the civil conflict and

the period of sanctions. Based on projected trends in prices globally, our projections suggest that there

will be modest continued increases in outputs for these three minerals over the coming decade. These

projections are heavily dependent on prices, which are consistently the most important predictor of

mining activity, so if future price projections increase or decrease, the future projections will change.



165

Table A2-8. Diamond mining outputs in Liberia, 1990-2025. Data is actual data from British Geological Survey and used as input

in the model. Prediction are the predicted values from the model estimated and presented in Figure A2-6.

Data Prediction

Year Carats/yr

1990 100,000 191,189

1991 100,000 174,402

1992 150,000 162,470

1993 150,000 126,588

1994 100,000 105,148

1995 150,000 87,428

1996 150,000 81,129

1997 200,000 84,915

1998 300,000 89,896

1999 200,000 101,503

2000 170,000 111,370

2001 155,000 105,756

2002 80,000 91,112

2003 60,000 56,013

2004 11,000 22,234

2005 11,000 24,299

2006 11,000 28,299

2007 21,699 46,024

2008 46,963 45,916

2009 36,828 43,361

2010 22,018 35,970

2011 39,866 42,994

2012 34,271 44,506

2013 44,334 44,982

2014 79,747 38,876

2015 41,636

2016 43,136

2017 44,691

2018 46,302

2019 47,971

2020 49,700

2021 51,492

2022 53,348



166

2023 55,271

2024 57,263

2025 59,327

Palm Oil Model

The final model developed for this analysis involves a demand and supply system for palm oil. Output in

Liberia has been fairly low since the 1960s and has only risen modestly over the time period (Figure A2-7).

Liberia reports fairly low levels of exports of palm oil as well, with Liberia only exporting 10% of total palm

oil production since 1966. This suggests that a significant proportion of palm oil in Liberia is consumed

locally.

Figure A2-7. Palm oil production in several African countries neighboring Liberia (UN FAO, 2016).

To model palm oil, we use the same demand and supply structure as used for the forestry sector above, although we use some different parameters. For prices, we use an instrument for the price of palm oil in Liberia, substituting the export price of palm oil from Ivory Coast. Ivory Coast produces and exports more



167

palm oil in general, and prices there potentially better represent the value of palm oil production in the region.

Supply: Ln(QSt )= α1 + α2Ln(Pt) + α3ERt+ α4Sanction + α5Ln(population) +et

Demand: Ln(QDt )= β1 + β2Ln(Pt) + β3Ln(USGDPPCt) + β4Ln(LiberiaGDPPCt) + β5Ln(FrImppricet) + β6Conflict + β7DC+ et

Equilibrium: Ln(QSt )= Ln(QDt)

Where:

Ln(QSt ): Natural log of palm production in tonnes (UN FAO, 2016) Ln(Pt): Natural log of palm export price in Ivory Coast (FUN AO,2016) Sanction: Dummy variable; 1 if during the period of UN sanctions (2003-2006), 0

otherwise. Ln(Populationt): Natural log of population aged 15 – 64 (World Bank, 2016) Ln(QDt ): Natural log of timber demanded in cubic meters (FAO, 2016) Ln(USGDPPCt): Natural log of Gross Domestic Product per capita in the US (World Bank,

2016) Ln(LiberiaGDPPCt): Natural log of Gross Domestic Product per capita in Liberia (World Bank,

2016) Ln(FrImppricet): Import price of palm oil in France (UN FAO, 2016) Conflict: Dummy variable for the period of civil conflict (1980-2003).

Model results are shown in Table A2-9. The parameters generally make sense with the price variable being negative in the demand function and positive in the supply function. The price parameter in the supply function, however, is not significantly different from 0 and the size of it is fairly small as well. This suggests that the supply function for palm oil in Liberia is not all that sensitive to prices and instead shifts in response to other factors, such as exchange rates and population. It is also likely that palm supply is a function of other factors for which we have not found the proper controls in this analysis, such as the price of land, and biological factors that control palm growth and output.



168

Table A2-9. Estimation results of palm oil supply and demand in Liberia (1966-2013)

Parameter Description Estimate Std. Error

α1 supply intercept -15.43*** 2.95

α2 palm export price 0.07 0.09

α3 official exchange rate -0.01*** 0

α4 sanction dummy 0.09 0.07

α5 Ln(Population) 1.83*** 0.19

β1 demand intercept -8.27*** 1.87

β2 palm export price -0.55** 0.21

β3 US GDP per capita (2005) 1.73*** 0.16

β4 Liberia GDP per capita (2005) 0*** 0

β5 palm import price in France (real) 0.59*** 0.18

β6 conflict dummy 0.12*** 0.04

Using the model in Table A2-9, future predicted palm output is shown in Figure A2-8, and the data for the prediction is provided in Table A2-10. Given relatively modest increases in income per capita in the US and Liberia, the model suggests that palm production increases over the next decade in Liberia, albeit at a relatively modest rate. This increase, while notable, would suggest that palm production in Liberia remains below that in Ivory Coast and Ghana, the two largest producing neighbors.



169

Figure A2-G8. Historical and predicted palm production in Liberia base on model estimation in Table A2-9.

Table A2-10. Palm oil outputs in Liberia, 1990-2025. Data is actual data from British Geological Survey and used as input in the

model. Prediction are the predicted values from the model estimated and presented in Figure A2-8

Data Prediction

Year Tonnes/yr

1990 23,500 31,643

1991 35,000 30,775

1992 35,000 30,418

1993 35,000 29,779

1994 36,000 29,968

1995 35,000 31,240

1996 32,915 32,678

1997 42,000 35,991

1998 42,000 41,487

1999 42,000 30,951

2000 42,000 33,912

2001 42,000 36,983

0

10000

20000

30000

40000

50000

60000

70000

1966 1976 1986 1996 2006 2016

Ton

nes

data Prediction



170

2002 42,000 36,708

2003 42,000 36,560

2004 42,000 38,318

2005 42,000 41,115

2006 42,000 42,221

2007 42,000 41,394

2008 42,000 43,210

2009 42,000 44,649

2010 42,000 44,976

2011 42,000 47,118

2012 43,500 49,117

2013 43,500 50,661

2014 51,075

2015 50,319

2016 51,347

2017 52,397

2018 53,469

2019 54,563

2020 55,679

2021 56,819

2022 57,983

2023 59,171

2024 60,384

2025 61,623

Discussion

The models provided in this analysis provide an initial look into market trends affecting important

extraction sectors in Liberia. Given the long-running civil conflict and the sanctions that were imposed in

the early 2000s, the results are surprisingly robust. We were able in all cases shown here, able to correlate

factors that would affect supply or demand with the output variables and/or prices. For timber and palm,

we developed demand and supply systems because there does appear to be an internal market for these

two products. For the mining sector we focused explicitly on identifying an output supply function,

assuming that there is little internal demand for the outputs of the mining sectors we analyzed.

The results suggest that outputs for most commodities in Liberia will be stable to increasing. For instance,

although wood products output is initially expected to fall in 2015, they are projected to increase after



171

that at a rate of 3.3% per year over the next decade. The reduction is largely due to the abeyance of the

current El Nino event. Although the parameter on the El Nino function is not significant, the parameter

has an important effect on the modeled outcomes. While some modelers would remove El Nino from the

analysis, it has important implications as an exogenous supply shifter and so is maintained. The longer

term increase in wood outputs relies largely on the assumed increases in GDP per capita outside of Liberia.

In contrast, gold outputs are projected to remain fairly constant. Gold prices increased substantially

through the 2000s, but these increases have abated recently. When looking at the projections our model

likely over-estimates the consequences of price changes in Liberia. That is, the high predicted output in

2012 is a function of the very high gold prices. The actual data does not indicate such a large increase in

gold output. Other factors that we have not modeled are also affecting gold outputs, but we have not

been able to include those in our analysis. We assume that gold prices remain constant in the next 10

years and this drives the moderation in outputs in our model.

Iron ore is projected to slowly rebound from the low period of no production from the early 1990s to 2010.

This slow increase is driven by the fact that we increase the dummy variable yy14 by 10% per year. While

this increase appears plausible given historical outputs from the 1970s and 1980s, it is important to

recognize that this is not tied to any assumed increases in exogenous factors like income. In fact, given

historical levels of output and the rapidity of the rise in output from 2010 to the present, it is plausible

that output could take another large step upward, particularly if new mines are opened. We are unable to

model the future eventuality of new mines opening with this analysis.

Diamond supplies also are expected to remain fairly constant in the future, increasing only modestly. The

main reason for this is that we assume that diamond prices are fairly stable in the future. We note that

there was a large increase in actual output in 2014, which our model does not capture because this

increase was not correlated with a large increase in prices. One issue to consider with the diamond model

is that our diamond price is for industrial diamonds, which have been commodified. In contrast, diamonds

used in the jewelry trade are extremely valuable, but have large variation in quality, which also affects

price. So this large increase in outputs could relate to specific discoveries of diamonds in markets outside

of the market for which we observe prices, namely industrial diamond markets.

Finally, we project a 1.7% per year increase in palm oil output in Liberia in the future. This is driven heavily

by rising prices for palm oil due to increasing demand elsewhere. Exports have not recently been a large

proportion of the total output in Liberia, but they do appear to be increasing. Our trends would include an

increase in palm oil exports. One of the limitations of this analysis is that we have not fully modeled the

supply side of Liberia, namely the biological components of production. These components would be

expected to have important impacts on future projected outputs.



172

The underlying data on outputs from the UN FAO, World Bank and the British Geological Survey represent

the most complete time series data we were able to find, however, there are clearly issues with the data.

These issues are not unique to Liberia, as they affect analysis in many countries of the world, but they

make inference such as produced in this report more complicated. For example, for the mining sectors,

we note that there are many years in the historical record when outputs are set to 0. While it is entirely

plausible that mining sectors started and stopped at various times, as appears to have happened in the

iron ore mining sector, it is also possible that reporting of mining was limited in the years with 0 reported

output. Also some mining outputs likely leave the country illegally. This may be most important with the

most valuable minerals that do not weigh as much, such as diamonds in particular, and gold.

Unfortunately, we can never know how much output is exported illegally, and we can never know if output

historically was truly 0. For the purposes of our analysis, we have assumed that 0's indicate 0 output.

There are also some concerns about the data for palm oil. The output data are flat from 1997 to 2011.

This is highly improbable given that outputs are a function of biological production functions. Palm

exports, not shown, displayed similar patterns of constant outputs for long periods of time, followed by

sudden changes. Although the aggregate output data are problematic, a potentially more important issue

in estimating the palm oil supply and demand system is that we lack a biological component in the supply

function. There could be a large number of factors influencing supply of palm that are related to annual

growth in the underlying resource. These factors include pests, drought, rainfall, etc. Importantly, we did

attempt to include an index for El Nino in our estimates, but it was not significant. While these biological

factors are important to consider, we note that the factors we have included, which control largely for

labor supply to harvest and use palm oil, are also critically important.

Despite concerns with the data, these results suggest that market factors are playing an important role

driving outputs in timber, palm oil, and mining in Liberia. The analyses could be improved with additional

effort and local data collection. First, for timber there are large fluctuations in the data that are not fully

captured by the model. It would be important to work with local partners to try to determine what factors

may actually have been at play in causing those fluctuations. If we could identify those and control for

them in our model, likely on the supply side but also potentially on the demand side, we could derive a

model with better fit for the timber analysis.

Second, an overall concern on the mining models relates to the role of individual mines and output. We

suspect that output is heavily influenced by individual mines in Liberia. It would be useful to determine

whether data would be available on individual mine openings and closings in Liberia. To obtain this data

for the complete historical record we have in this analysis would likely be exceedingly difficult to do for

gold and diamonds, and potentially iron ore. Assessing whether this data is available, however, could

provide better options for making projections.



173

Finally, with respect to palm oil production, aside from data, the biggest limitation in the analysis is the

lack of a biological control on the supply side. Palm oil is a renewable resource and production of palm oil

is tied to growth in the palm resource. Better weather data could be explored to control for the historical

influence of weather on palm oil production. Other known large scale events could also be controlled if

local experts can provide such information. It would also be useful to consider an alternative type of

modeling of palm oil development for Liberia based on structural dynamic methods that directly model

the underlying palm oil resource base. Such methods could be developed with the relatively scarce data

that already exist.

One component of this analysis that did not work out was estimates of factors influencing actual land use

change. Using land cover change data developed from 2001-2014, we attempted to correlate mining,

forestry, and palm prices to deforestation rates, but were unsuccessful. Population, not surprisingly, was

correlated with deforestation rates. We did not incorporate agricultural prices in that analysis, but

agricultural prices likely would play a role. One reason for the difficulty in making the link between prices

and deforestation, we suspect, has to do with needing to tie deforestation to specific activities, e.g.,

mining, palm development, agriculture. If data on the drivers of deforestation were available over time,

we could attempt to develop land use change models.

References

Blundell, A., Callamand, D., Carisch, E., Fithen, C., and Kelley, H. 2003 “Report of the Panel of Experts pursuant to paragraph 25 of Security Council resolution 1478 (2003) concerning Liberia.” United Nations Security Council, Document number: S/2003/779. Blundell, A., Callamand, D., Fithen, C., Garnett, T., and Sinha, RH. 2005 “Report of the Panel of Experts, submitted pursuant to paragraph 14 (e) of Security Council resolution 1607 (2005) concerning Liberia.” United Nations Security Council, Document number: S/2005/745. Blundell, A., Callamand, D., Fithen, C., Garnett, T., and Sinha, RH. 2007 “Report of the Panel of Experts submitted pursuant to paragraph 4 (d) of Security Council resolution 1731 (2006) concerning Liberia.” United Nations Security Council, Document number: S/2007/340. British Geological Survey. 2016. World Mineral Production. http://www.bgs.ac.uk/mineralsuk/statistics/worldStatistics.html

http://www.bgs.ac.uk/mineralsuk/statistics/worldStatistics.html



174

Halton, A. 2013 “Background: A Snapshot of Forest Governance Assessment for REDD+ in Liberia” in

Liberia: Assessment of key governance issues for REDD+ implementation through application of the

PROFOR forest governance tool by Forest Development Authority (FDA), Liberia.

U.S. Geological Survey. 2014. Historical statistics for mineral and material commodities in the United

States: U.S. Geological Survey Data Series 140. Kelly, T.D., and Matos, G.R. (eds).

http://minerals.usgs.gov/minerals/pubs/historical-statistics/.

UN FAO. 2016. FAOSTATS database. www.fao.org

World Bank. 2016. http://data.worldbank.org/

http://minerals.usgs.gov/minerals/pubs/historical-statistics/

http://www.fao.org/

http://data.worldbank.org/



175

ANNEX 3: CAPACITY BUILDING STRATEGY An important component of the development of recommendations for a REDD+ Reference Scenario for

Liberia is capacity building to assist the country in developing the skills and knowledge to understand,

maintain, and revise the reference scenario within country, to the extent possible. An initial capacity

assessment has been conducted through meetings with the relevant entities and review of existing

documents, notably Liberia’s REDD Readiness Preparation Proposal (described below). While there are

certain individuals who have a strong understanding of REDD+ in general, the country currently has limited

data on forest biomass and land use, limited technological capacity, particularly in regards to spatial

analysis, and limited overall knowledge about REDD+ Reference Levels and their development.

Capacity building activities take many forms, including reports such as this one, which describe the

elements of a reference level and the decision points needed; meetings to discuss such decisions and

provide additional description and guidance; workshops in which broad components can be discussed,

with broad participant; and technical trainings, aimed at specific technical staff who will implement future

reference level and MRV activities. This annex provides a capacity assessment for Liberia’s REDD+

Reference Level knowledge and expertise, detail on the capacity building activities that have been

undertaken to date, and a description of additional capacity building needs that are recommended for the

future. At the end of this annex, the capacity building activities described in the MRV Roadmap of June

2016 are linked to the capacity building strategy described here.

Capacity assessment report

Recognizing the overall capacity gaps and need that exist regarding REDD+ in Liberia, WI/CI in consultation

with the REDD+ Implementation Unit (RIU) design a REDD+ capacity assessment questionnaire template

for the purpose of collecting relevant data to be in as part of Liberia’s reference level development

capacity building processes. The designed capacity assessment questionnaire template was divided into

four main categories, which are summarized below:

I. General REDD+ Knowledge on the organization past and present REDD+ activities, plans for

future REDD+ engagement, and level of knowledge and expertise on international UNFCCC

processes on REDD and IPCC LULUCF monitoring requirements

II. Remote sensing and GIS expertise within the organization the number of expertise, training

levels and years of experience in this area. This section also focuses on collecting information

of the kind of GIS software and used satellite imagery currently used by the organization and

the number of staff with the appropriate expertise.

III. Forest inventory expertise within the organization the level of expertise that exists in forest

inventory, the training capacity of available staff and the years of experience in the area. This



176

section also focuses on identifying the kinds of equipment the organization currently used to

conduct forest inventory and the types of forest inventories work carryout. Another

information collected about organization is the level of expertise in the used of data analysis

software.

IV. Stakeholder engagement: this section focuses on levels of efforts undertaken by organization

to engage stakeholders or increase stakeholders participation and understanding about their

program and project activities. It simply target understanding the most frequently methods

used by organization for the purpose of engaging stakeholders.

The REDD+ Implementation Units recommended two key government institutions GIS/ RS departments to

be assessed under the current national Reference levels development process, which included the

Forestry Development Authority and Liberia Institute of Statistics and Geo-Information Services (LISGIS).

The capacity assessment data were collected through the administered questionaire to the head of the

GIS/ RS department in each institution. Below are summaries key findings:

I. The Forestry development Authority (FDA):

The department of research and development is responsible GIS/ RS in the FDA. The department

demonstrated involvement in GIS/ RS training and data collection exercise facilitated by

Metria/Geoville, and participated in REDD+ training and carbon measurement demonstration

training at the Wonegizi REDD+ pilot site organized by Forest Trends and Conservation

International. No staff within the department have had training or knowledge in UNFCCC

processes on REDD and IPCC LULUCF monitoring requirements.

The research and development department has a GIS Laboratory that oversees all remote sensing,

GIS and forest inventory work. There are currently 13 staff in this lab with 5 staff assigned at the

central office and 8 staff assigned at the four FDA regional offices. Within staff that have full

employment, 2 staff hold bachelor degree in general forestry, 2 staff hold certificate in GIS/ RS

from Ghana and India, 1 staff holds a post- graduate certificate from Nigeria. The average staff

working experiences ranges between 1- 8 years respectfully. However, 1 staff holds a certificate

in drafting and is being used as the department’s cartographer, and has 27 years working

experience. The GIS Lab does not have a valid GIS/ RS license software, staff use open source

software that were installed by the VPA project. However, there are staff that have training in the

use of several GIS software including QGIS, IGIS, ARCVIEW, ARCGIS, IDRISI, EDRAS and ENVI. The

department used satellite imagery for its work during the leadership of Mr. Augustin Johnson, but

not at present.

There are 8 staffs with expertise in forest inventory, 2 staffs have a bachelor’s degree in general

forestry and 4 staff hold a certificate in forest inventory training conducted by Conservation

International (Peter Herbs- trainer). The experience of staff in forestry inventories ranges between

6- 11 years. Reports show that the department lack the requirements and necessary equipment

needed to conduct forest inventory, nevertheless, staff demonstrated knowledge of the use of



177

GPS Units ( Garmin Max 62), Clinometer, 50 meter tape, diameter tape and declinometer from

previous work. In previous years, the department have had expertise in the use of forest inventory

methods like Simple random sampling, cluster plots, Forest dense sampling and Agriculture

degraded sampling. Staffs also have beginner level of expertise in the used of Microsoft Excel as

the data analysis software. The department of research and development does not have

stakeholders’ engagement and consultation as part of its mandate. The role of the GIS Lab is to

produce and interpret GIS/ RS information and maps.

II. Liberia Institute of Statistics and Geo-Information Services (LISGIS)

The mandate of LISGIS is to compile statistical data on the status of demographic and other socioeconomic indicators and to coordinate the dissemination of official statistics on Liberia. LISGIS involvement in REDD+ activities had been through national land cover and land- use suitability studies conducted by the Land Commission and participation in the ground trothing and plot verification of sample plots produce by Metria/Geoville. No staff within the department have had training or knowledge in UNFCCC processes on REDD and IPCC LULUCF monitoring requirements.

The department of Geo-Information services and coordination at LISGIS is responsible for acquisition of spatial data and production of spatial information products and producing and disseminating publications/documentations and maps. This department has two main divisions with assigned staffing.

Laboratory

The laboratory has 10 staff, 2 staff holds certificates in GIS from Ghana and Nigeria, while the rest of staff

hold Bachelor degree in none GIS related disciplines. The average staffs work experiences range between

4- 8 years.

Cartography

The cartography division has 5 staff which are assigned at the central office and 1 staff that holds a

certificate in cartography from the Netherlands. The remaining 4 staff hold Bachelor degrees in none GIS

related disciplines.

Beside staff that are assigned at LISGIS central office, there are 2 staff assigned in each of the 15

counties (30 staffs) which are mainly beginners. These staffs have capacity in using GPS, camera,

laptop and A3 map production printer. Information collected demonstrates that most staff have

experience in the use of GIS software like ArcGIS (Arcinfo & ArcView inclusive), and QGIS (free

version 7). One staff proves to have experiences in ERDAS, Idrisi and Mapinfo (version 2). LISGIS

has experiences in the use of satellite imagery like Landsat- for forest cover, Spot- human

settlement mapping, and Ikonos- urban planning. LISGIS do not have separate staff for forest



178

inventory, however, three staff are been designated to work with FDA on the REDD+ forest

inventory activities after the Metria/Geoville training. All staffs are knowledgeable in the used of

forest inventory equipment like GPS units (Garmin, spectra & Tremble), Diameter tape, Tablet/

mobile phone and Clustered plots. One staff has expertise in the use of data analysis software like

Microsoft Excel, Microsoft Access, STATA, SPSS and CS pro.

LISGIS disseminates data and information to stakeholders through the use of the following tools-

CensusInfo, LiberiaInfo, Liberia National Data Archive, DevInfo, IMIS, Liberia Data Portal and Open

Data for Africa.

III. Capacity building activities conducted

1. Technical Training on RL Development

Between 16-21 April 2016, Winrock International and Conservation International facilitated a four days technical training on reference levels development which focuses on the technical aspects of reference level creation. The aim of the training was to strengthen the capacity within Liberia to establish and maintain a REDD+ Reference Level. Over 40 technical staff with knowledge and experience with forest inventory and/or GIS and remote sensing participated in this training from key national institutions including LISGIS, FDA, SCNL, Green Advocate, Ministry of Lands, Mines and Energy, Land Commission, National Bureau of Concessions. Key topics taught during this training workshop include the followings:

1. Overview of Climate Change, UNFCCC, National REDD+ framework, and the Importance of

RLs

a. IPCC activity accounting methods

b. Summary of RL and MRV creation

c. Current Status of Liberia’s activities in REDD+ readiness preparation

d. Overview on Liberia Land Cover and Forest Mapping project and its specific

contribution to REDD+.

e. Presentation of the Liberia Land Cover and Forest Map 2015

2. Historical Emissions Analysis a. Key Decisions in National REDD+ Mechanism Development b. Key Decisions in National REDD+ Mechanism Development c. Overview of development of draft Reference Level d. Land cover change mapping e. Developing Activity Data for Deforestation 21.

3. Historical Emissions Analysis and RL development

a. Focused on technical issues

b. Hands-on Exercise: Satellite Image Analysis for Land Cover Change



179

c. Land Cover Carbon Stock Estimation

d. Hands-on Exercise: Using existing data for use in carbon estimation

e. Forest Carbon Stratification Techniques

f. Carbon Stock Sampling Design & Plot Distribution

g. Carbon Stock Sampling Design, Field Measurements, and Analysis

4. RL development and next steps

a. Hands-on Exercise: Carbon Stock Estimation Creation

b. Create Deforestation Emission Factors

c. Combining Activity Data and Emissions Factors

d. Hands-on Exercise: Historical Emissions - Bringing the components together

e. Moving from Historical Emissions to RL

f. Hands-on Exercise: RL Creation

Capacity Building efforts undertaken within Reference Level

development project

Introductory training on reference levels

In September 2015, an initial meeting was held with the REDD Technical Working Group and staff from

Winrock International and Conservation International Liberia. During this meeting, a presentation was

given describing what a reference level is, and what is required for a country to develop a REDD+ Reference

Level. Guidance from the UNFCCC, the Carbon Fund, and the IPCC as relates to reference level

development was provided; the basic decisions that must be made were explained; and an overview of

the technical components was given.

Forest Definition Workshop

In January 2016, FDA sponsored a Forest Definition Workshop, during which Winrock International

provided key presentations on forest definition and reference levels: 1) Overview of a REDD+ Forest

Definition, 2) Options for Liberia’s Forest Definition, and 3) Overview of development of a Reference Level

for Liberia. The first presentation provided a general understanding of the components of a forest

definition and how this is addressed in the REDD+ context. The second presentation described the

implications of various definitions in the context of Liberia. These set the stage for breakout sessions held

during the workshop, and ultimately for the adoption of a forest definition for Liberia.



180

In February 2016, following the Forest Definition Workshop, a second meeting was held with the RTWG.

During this meeting, the status of reference level development was described, with particular attention to

the critical decisions that must be made by Liberia and what the implications are for various options.

Additionally, the analysis done to date (and described in detail in this report) was described, with the

intention of clarifying any questions and assisting in finalizing the necessary decisions.

Technical Training on REL Development

In April 2016, Winrock provided a week long training for technical staff in Liberia to gain an understanding

about REL creation and maintenance, using suitable country-specific data related to activity data and

emission factors. A key to the success of this training was the coordinated interaction with the appropriate

Government agencies (at the national and sub-national levels) to provide necessary policy guidance and

direction to ensure that REL development is in line with the central government regulations and provincial

socio-economic development targets.

The aim of the training was to strengthen the capacity within Liberia to establish and maintain a REDD+

Reference Emission Level. By the end of the training participants were expected to:

Gain an understanding of the components of REL creation

Participate in a hands on example of REL creation

Understand technical steps required to create REL

Know the types of data that must be developed to create activity data and emission factors

Understand the implications of using global, regional, and country-specific data.

This training was geared to the technical staff who have knowledge of and experience with forest

inventory and/or GIS and remote sensing, both of which are necessary for the development of a Reference

Level.

This training only focused on the technical aspects of reference level creation. It did not include training

on stakeholder engagement, policy development, or REDD+ strategy creation. It also did not

comprehensively address field data collection and analysis required to estimate carbon stocks, though

these topics were introduced.

Additional Capacity Building Needs

The work conducted above along with other workshops and trainings undertaken in Liberia on REDD+

readiness have introduced the concepts necessary to develop an REL and MRV mechanism. However, in

order to develop the in-country capacity to develop a reference level, additional capacity building will be



181

required in Liberia, so that certain technical gaps can be filled and country representatives can provide the

necessary oversight where additional support from outside sources will be needed. In addition to targeted

capacity building, supporting documentation that provides stepwise guidance and standard operating

procedures are necessary to ensure that the methods developed can be replicated in the future.

This section highlights recommended capacity building efforts and supporting documentation geared

towards the RTWG, the RIU, policy makers, and technical GIS and forest inventory staff, so that

improvements can be made to the recommended reference level. Table A3-1 provides a listing of

recommended trainings, with additional descriptions in the text below.

Table A3-1. Training needs in order of importance.

Training/Supporting Documentation

Importance Justification

Forest Carbon Stock Development High Liberia must develop its own carbon stocks to develop emission factors for deforestation

Reference Level Development High This training is essential to consolidate the work done to date and use country-specific data to develop the RL

Geospatial Training High It is important to understand the use of geospatial products to assess forest cover change and what products can be used to do so.

Standard Operating procedures High It is necessary to develop standard operating procedures so that work can be replicated in the future using the same methods.

Emission Calculation Tools High Tools provide a framework for emission calculations, so that these can systematically be replicated as data is updated.

Destructive Sampling for Allometric Equation Verification/Development

Medium It is important for Liberia to verify whether its forest can use existing allometric equation, but this training will depend on resources available.

Guidance Documents Medium Guidance documents tailored for Liberia would provide information on how to develop the different aspects of the RL and MRV (e.g. emission factors for deforestation, activity data for deforestation etc.)

Developing Emission Factors for Logging

Low The focus in first stage should be on deforestation. If degradation is assessed a decision must be made on whether Liberia follows and landscape based approach or an activity based approach. If an activity based approach is chosen and emissions from legal logging are considered significant, this training will be important.



182

Biomass Data and Analysis to establish Emission factors

This training is the highest priority of the trainings outlined here as it is essential for Liberia to develop Tier

3 country specific forest carbon stocks. As stated in this report, Liberia currently lacks the necessary forest

inventory data needed to develop such emission factors for its REDD+ program. Liberia would benefit from

a targeted training on forest carbon inventory development. This training would consist of a field

component and a classroom component for forest inventory staff.

For the field component, participants should be trained on the field measurements necessary to assess

carbon stock in forest each carbon pool (see Figure A3-1) for the strata identified in the 2000 land cover

classification and the current forest classification.

Figure A3-1. Forest carbon pools and definition of an emission factor for deforestation.

Directly following the field work, participants should be trained in the analysis of the carbons stocks

measured. This analysis should include a lab component, for sample analysis, and a hands-on classroom

component where participants are taught how to derive carbon stock estimates per forest strata based

on the measurements taken.

Training on geospatial products used for REDD+



183

This training will focus on the LISGIS staff and the FDA research and development department’s GIS

Laboratory. The goals are (1) to introduce the basis of remote sensing (RS), available RS platforms, RS

techniques for image procession and interpretation for the forestry sector, the skills and technical

requirements for implementation of the RS into the forest change monitoring, (2) to provide opportunity

for the participants to experience the use of this technology through hands-on practices and (3) to present

a broad brush study on the use of global land cover datasets to quantify the CO2 emissions from forestry

sector.

Training on destructive sampling for allometric equation verification (and development if

needed)

As noted in this report, Liberia does not have specific allometric equations to estimate biomass from basic

field measurements. Allometric equations are used commonly to estimate tree biomass from

measurements of DBH or DBH and height. Different equations give different estimates for tree biomass

because each is designed for a specific forest and climate type. Before applying a regression equation to

estimate forest biomass, it is necessary to destructively sample a few trees to check the appropriateness

of an existing equation or, if no equation is found to be appropriate, develop a new equation.

A training to conduct this work would also consist of two steps. Field work focused on destructive sampling

and a hands on classroom session focused on data entry and analysis.

The field teams would to visit logging blocks where felled trees remain on site (i.e. before logs have been

hauled away). This will avoid the need to cut additional large trees for the sole purpose of destructive

harvest measurements. To verify the applicability of the selected biomass equation, trees of a range of

DBH (small, medium, large trees) should be measured. Trees that were cut previously for timber will be

measured, and smaller trees can be cut and measured in unmanaged areas so that no new tree felling will

occur within logging blocks without prior permission from the concession.

Following field training and data collection, class room based sessions will strengthen the capacity of

participants in laboratory procedures, and data entry and analysis required to transform field

measurements into meaningful information to be used in forest carbon monitoring systems. This aspect

of training is highly recommended as it ensures that participants understand the treatment of data

collected in the field, how it translates into meaningful parameters, and the importance of precision and

accuracy of data collection when in the field.



184

Training on emissions from logging

With logging as the largest estimated source of GHG emission from degradation, Liberia should assess

emissions from this activity. The objective of this training would be to support Liberia in the measurements

and data analysis necessary to estimate emissions from forest degradation from logging. As with the other

trainings it would consist of a field component and a classroom component.

Participants will be trained in the field on how to take the measurements necessary to develop emission

factors from selective logging. In the classroom participants would use a tool assess the emissions

associated with logging damage, logging infrastructure and the emissions from the actual log.

It is important to note that emission factors developed using this method require reliable estimates of

timber volume harvested.

Annex 1 of this document includes more explicit steps for developing emissions estimates for selective

logging, as well as other activities that result in forest degradation.

Follow up hands-on training on Reference Level

Once sufficient data are collected to estimate the reference level with Liberia specific emission factors for

the activities under Liberia’s RL, a follow up hands-on training on RL establishment. The goals of this

training workshop are to build an understanding of the components of RL creation, and related data and

analysis requirements using data developed for Liberia. The workshop will be designed to include a

mixture of presentations and hands-on exercises with geospatial and carbon data. Topics that will be

covered in this training workshop include:

REDD+ Review

Reference Level Development Planning

Historical Emissions Overview

Historical Deforestation Emissions

Historical Degradation Emissions

Creation a RL

The training will be an opportunity for the technical staff at national and regional level to gain

understandings about RL creation as well as being able to participate in hands-on exercises for developing

a RL considering Liberia-specific circumstances and using appropriate data and methods. One major

outcome from this training should be the development of a reference level using a tool that Liberia will

use in its MRV.



185

Provide SOPs and Computing Tools Tailored for Liberia

All data collection must be conducted following SOPs that specifically speak to the realities of the forest

in Liberia. These SOPs must provide step by step instructions on the data that needs to be collected to

complete a full forest carbon inventory, destructive sampling approach and logging field measurements

together with quality assurance and quality control procedures.

Tools for data entry and analysis should also be developed in order to streamline data analysis for each

component. Both the SOP and the computing accounting tools for logging emissions, carbon stock

assessment and destructive sampling will help to achieve quality assurance and control in the

development of deforestation emission factors.

Guidance Documents

Liberia would benefit from a suite of guidance documents outlining the development of National Forest

monitoring system for REDD+ (NFMS), which provides information on the technical requirements for

establishing a NFMS to produce the data and information inputs that will be used to establish the RL/REL

and that will feed into the MRV system. The guidance series should be divided into multiple modules

describing different steps and technical components required to establish the NFMS and estimate

historical emissions to develop the RL/REL. Each module would describe, in a step-wise manner, the good

practice guidance needed to produce transparent, complete, comparable, and consistent estimates of

gross and net emissions with low uncertainties based on use of the Intergovernmental Panel on Climate

Change (IPCC) framework methodologies and tailored for Liberia.

186 VERSIO N : APRIL 2016 – W INROC K TERRE STRI AL SE LECT IVE LOG GI NG SOP

Winrock International 2016

Key Actions from MRV Roadmap

The MRV roadmap, finalized June 2016, provides a section on “key activities for capacity development.” In that section, seven themes are addressed, some of which have an administrative focus (i.e. “establish institutional arrangements”) and some of which have a technical focus (i.e. “improve national forest monitoring”). It is important to note that many of the items described in the MRV roadmap have been completed, or are in process, as part of Liberia’s ongoing REDD+ activities, including the REL/RL development. Here is provided the summary of key actions from the MRV roadmap, with additional notes related to capacity building activities discussed in the strategy above. Activity Responsible agencies Other stakeholders and

potential

(international) partners

Timeframe Link to RL/REL Capacity

Building Strategy

1. Establish institutional arrangements

1.1 Establish

steering/coordination body for

the REDD+ NFMS/MRV

system

EPA, FDA, MFDP LISGIS, Ministries,

CSOs, INGOs, WB

FCPF, RSPB/SCNL,

donors

Immediate (within first 6

months)

Recommend this be highly

engaged subset of RIU/RTWG

1.2 Establish technical working

group(s) and facilities within

FDA and with partners

FDA, EPA EPA, LISGIS, MoA /

MoGD / MIA / MoPEA

/ MLME, RSPB/SCNL,

CSOs,

Short-term (within first year) Highly engaged TWG needed

1.3 Establish a mechanism for

local engagement and exchange

of capacities, experiences and

data between national and local

forest monitoring activities

FDA CSOs, RSPB/SCNL,

Communities, NGOs

Short-term Critical for further developing

in country expertise and

capacity

1.4 Develop a framework to

engage with research and higher

education institutions

FDA Universities, research

institutions, LISGIS,

WRI, CI, EPA,

RSPB/SCNL

Short-term Critical for further developing

in country expertise and

capacity

2. Improve national forest monitoring: activity data

2.1 Decide on a forest definition FDA, EPA FAO/UN-REDD, WB

FCPF, Wageningen

University

Immediate Stakeholder workshop held in

January 2016, with definition

approved by FDA MD



2.2 Update and improve

national forest map and/or land

use map

FDA, LISGIS International

consultants, FAO/UN-

REDD and other

institutes for training

Short-term 2014 forest cover map

developed by Metria Geoville;

recommend it be updated

annually if possible; crucial

improvements include detailed

land use mapping (esp. active

plantation and timber harvest)

2.3 Estimate changes in forest

area at national level



REDD and other


Short-term; recurrent Annual forest change has been

mapped for 2000-2014. Should

continue into future, with

stepwise improvement (e.g.

inclusion of detailed land use

and addition of degradation

monitoring)

2.4 Estimate activity data for

forest degradation



REDD and other


Medium-term (within 2-3

years); recurrent

This will require improved data

collection as described in

Annex 1.

2.5 Estimate activity data for

enhancement, sustainable

management of forests and/or

conservation



REDD and other


Medium-term; recurrent This will require improved data

collection and is viewed as a

longer term objective

3. Improve national forest monitoring: carbon stocks and emission factors

3.1 Design/update and

implement a national forest

inventory and carbon

measurement system

FDA FAO/UN-REDD,

RSPB/SCNL,

consultancies/CSOs

Short-term Critical for establishing Tier 1

emission factors. Additional

detail provided in “Guidance on

Developing a National Forest

Inventory for Forest Carbon

Sampling”

3.2 Develop factors for: Carbon

Conversion, Expansion Factors,

Wood Density and Root/Shoot

Ratio, and convert existing

forest and forestry data into

carbon

FDA, Universities,

research institutions

FAO/UN-REDD,

RSPB/SCNL,

consultancies

Medium-term Needed factors depend on

design of NFMS; some factors

recommended in RL report

(e.g. R/S ratio)



3.3 Assess different

drivers/processes of change and

their carbon impact in order to

develop emission factors

FDA FAO/UN-REDD,

RSPB/SCNL,

consultancies/CSOs

Medium-term Initial driver analyses have

been conducted (see e.g. R-PP,

REDD+ Strategy); these could

be improved and expanded

4. Improve estimation and international LULUCF, GHG inventory and REDD+ reporting capacities

4.1 Engage in technical support

and training for national GHG

inventories and for upcoming

REDD+ reporting

EPA RIU, LISGIS,

International partners

(WRI, CI, FFI, VPA),

RSPB/SCNL, World

Bank FCPF

Immediate Critical ongoing effort

4.2. Assess historical national

GHG inventories for the

LULUCF/AFOLU sector,

appraise gaps and needs for

alignment in the context of

REDD+ and ensure

streamlining of REDD+ and

GHG reporting in National

Communications and Biennial

Update Reports

EPA RIU, LISGIS,



RSPB/SCNL

Short-term Critical ongoing effort

4.3 Decide on a forest reference

emission level (FRL/FREL),

which is based on historical

data and adjusted for national

circumstances

FDA, EPA World Bank FCPF,

Universities, research

institutions, Winrock

International, FAO/UN-

REDD, RSPB/SCNL

Short-term Initial FREL described in

current report, with

recommended step-wise

improvement detailed

4.4 Develop technical annex of

the BUR, to make the REDD+

results available for technical

assessment, in the context of

results-based payments

EPA RIU, LISGIS,



RSPB/SCNL

Medium-term

4.5 Ensure that the data

collected in the context of the

MRV system or NFMS is used

in the above exercises for the

LULUCF sector to ensure

consistency between the GHGs

EPA RIU, LISGIS,



RSPB/SCNL

Short-term



inventory the BUR annex for

REDD+ and the reference level

4.6 Submit an “MRVS interim

measures report”, including the

FRL and performance reporting

for the target landscapes

EPA RIU, LISGIS,



RSPB/SCNL

Short-term (end 2016) MRV should be conducted in

accordance with methods used

for FREL/FRL

4.7 Submit an “MRVS interim

measures report”, including the

FRL and performance reporting

for the whole country as

baseline and model for

continued performance

reporting

EPA RIU, LISGIS,



RSPB/SCNL

Medium-term (mid 2017) MRV should be conducted in

accordance with methods used

for FREL/FRL

5. Prepare for MRV of REDD+ activities on the national level

5.1 Adapt and develop the

national forest monitoring for

local/landscape-scale REDD+

demonstration activities

FDA World Bank FCPF,

LISGIS, NGOs,

communities,

RSPB/SCNL, WRI,

FAO/UN-REDD

Short-term Nesting should follow best

practices; see for instance,

http://www.v-c-s.org/wp-

content/uploads/2016/07/Nestin

g-Options-1-Jul_Eng_final.pdf

5.2 Test approaches and options

to derive forest reference

(emission) levels


Universities, research

institutions, Winrock

International, FAO/UN-

REDD

Short-term Described within current REL

report

5.3 Develop foundations and

data sources for a REDD+

safeguard information system


RSPB/SCNL, NGOs,

FAO/UN-REDD

Medium-term

6. Implement a program for continuous improvement and capacity development

6.1 Design and implement a

capacity development program

building on available national

capacities and international

support where needed


FAO/UN-REDD,

GOFC- GOLD, GFOI,

WRI, Silvacarbon,

RSPB/SCNL

Short-term to Medium-term






6.2 Implement a program to

“train the trainers” and to

multiply capacities within the

country, perhaps establish a

dedicated training unit for

REDD+ monitoring aiming for

sustainable training capacities


FAO/UN-REDD,

GOFC-GOLD, GFOI,

Silvacarbon,

RSPB/SCNL

Short-term to Medium-term

6.3 Establish a team of

(international) experts that can

serve for backstopping and

advisory group for key

decisions to be made


FAO/UN-REDD,


(Winrock International,

WRI, CI, FFI, VPA),

RSPB/SCNL

Medium-term

6.4 Seek partnerships with

regional organizations and

international partners (i.e.

South-South cooperation and

student/staff exchange etc.)

FDA, EPA Universities, research

institutions, FAO/UN-

REDD, International

partners (Winrock

International, WRI, CI,

FFI, VPA), RSPB/SCNL

Medium-term

7. Continued national and local communication mechanism on REDD+ monitoring

7.1 Conduct a series of regional

workshops to inform about

REDD+ and MRV among

national, regional and local

actors, in order to both inform

and seek input from different

stakeholders; in particular

involving local communities

FDA, EPA Communities,

NGO/CSOs

Short-term to Medium term

7.2 Produce communication

plan, communication materials

on REDD+ and monitoring

FDA, EPA NGOs, FAO/UN-

REDD, WB FCPF

Short-term

7.3 Establish and maintain

REDD+ monitoring web-site

with relevant information and

outreach materials

FDA, EPA NGOs Short-term

191 G C C - C R I T A S K 2 : A P P R O A C H A N D I N P U T T O T O O L S

Dr. Sarah M Walker

Program Director

Ecosystem Services,

Winrock International

[email protected]

office +1.703.302.6556 2121 Crystal Drive, Suite 500

Arlington, VA 22202, USA www.winrock.org

forestcarbonpartnership.org · LIBERIA REDD+ REFERENCE SCENARIO DELIVERABLE 4 – FINAL REPORT 2...

Documents

Transcript of forestcarbonpartnership.org · LIBERIA REDD+ REFERENCE SCENARIO DELIVERABLE 4 – FINAL REPORT 2...